Blaming ProtonMail misses important lessons of the case, as request from authorities ticked the necessary requirements under Swiss law

ProtonMail, a privacy and security-focused email provider based in Switzerland, has been strongly criticized for providing the IP address of a customer to Swiss authorities, ultimately leading to the arrest of a climate activist in France. But simply blaming ProtonMail misses the important lessons of this case.

Background

French authorities were aware that a group ‘of interest’ (the Youth for Climate collective and associated groups) used the jmm18[@]protonmail.com email address. According to police reports, the climate group had hardened its interests along general anti-capitalist lines, and were taking part in illegal squatting and damage to property.

Since Switzerland is not part of the EU, the French police could not demand that the Swiss authorities obtain and hand over the IP address of the email user. Instead, it approached Switzerland via Europol. Switzerland acquiesced with Europol, and required ProtonMail to deliver up the IP address. Since the request ticked all the necessary requirements under Swiss law, ProtonMail had no option but to obey.

It should be stressed that ProtonMail cannot deliver the content of its end-to-end encryption – this is solely about the user’s IP address.

ProtonMail

ProtonMail is not happy with the events. It published a blog titled Important clarifications regarding arrest of climate activist on September 6, 2021, commenting, “We are also deeply concerned about this case and deplore that the legal tools for serious crimes are being used in this way.”

Several points stand out in this blog – most importantly that ProtonMail had no alternative but to comply with the Swiss court order. ProtonMail does not know the identity of its users. “We only know that the order for data from the Swiss government came through channels typically reserved for serious crimes.” It did not know that the target of observation was a group of French climate activists – for all ProtonMail knew, it could have been a gang of international terrorists.

Noticeably, the blog provides no information on the details of the case. However, it directs readers to the ProtonMail transparency report, and the ProtonMail privacy policy. The former states, “ProtonMail may also be obligated to monitor the IP addresses which are being used to access the ProtonMail accounts which are engaged in criminal activities.” The latter includes, “If permitted by law, we will always contact a user first before any data disclosure.”

It is possible that ProtonMail was required to monitor the IP or IPs used by the email address over a period of time, while simultaneously under a gagging order not to disclose the fact. Only after an arrest could matters be made public.

Lessons to be learned

SecurityWeek talked to European privacy activist and advocate Peter Sunde Kolmisoppi. He works on projects designed to help people protect their privacy (such as Njal.la, a privacy aware domain service with both VPS and VPN), but also encourages groups and political entities on increasing the right to privacy. He is a co-founder of The Pirate Bay and the founder of Flattr and Njal.la.

Sunde does not blame ProtonMail for what happened, but that doesn’t mean he is not angry at what did happen. The primary problem, he suggests, is that ProtonMail is “based in a country that has a government that can control their actions. That’s the main flaw. A lot of activists, technologists and hosters have this idea that certain countries are ‘bulletproof’ when it comes to privacy. That’s certainly not the case.”

This is the first lesson to be learned – to recognize that all companies are subject to the laws of the country where they reside, irrespective of their own principles and preferences.

He believes the solution here would be to decentralize the service – something eminently feasible given todays’ global public cloud. “The basic problem here is again that we’ve centralized things. Organizations, services – e-mail is among the easiest of all services to have decentralized – and trust.” But ProtonMail is what it is, and alternative ideas are irrelevant to the current situation.

“It’s not the fault of ProtonMail, he says, it’s the fault of the authorities. And I’m sure that ProtonMail will take lessons from this to improve their threat model.” ProtonMail has started to stress the availability of its onion service to allow users with heightened threat conditions to gain the additional protection of Tor, and provides clear access to its own VPN, ProtonVPN, on the home page.

Swiss law treats VPN different to email. “Under current Swiss law, email and VPN are treated differently, and ProtonVPN cannot be compelled to log user data,” wrote ProtonMail in its blog. The implication is that the French authorities may not have been able to secure the IP address had the activists been using ProtonMail with Tor and VPN as well as ProtonMail.

This is the second lesson – when using any service, be aware of the additional security protection that may be available, and use it.

Sunde’s anger is directed against the authorities rather than ProtonMail. It seems that the route taken to get the IP address was originally intended for just the most serious of crimes – not to be used against climate activists (although in fairness we don’t really know what else the authorities were hoping or expecting to find).

“Switzerland is not part of the EU, so we’re not sure why Switzerland nor the EU looked at this as a case as important enough to go to these extreme measures,” he told SecurityWeek. “There should have been many people that could have stopped this insanity.”

He continued with the third lesson to be learned: “In general, I would say that people should not trust any single entity/provider to protect any secret communication, because most of these organizations can be forced to do things, or there can be backdoors or security issues with other things. An app for secure chat is great but it will ultimately come down to if your computer/phone is secure – and it is not.”

Finally, wearing his activist hat, he added, “I think we need to shame both the Swiss and EU authorities for allowing this ridiculous thing to happen; and hopefully this situation will lead to better decisions on what they will do in the future.”

It is worth also adding that although there is no direct comparison, the primary argument that has led the European courts to issue the Schrems II ruling has been the ability of the NSA to obtain European personal information. In this instance, the French authorities obtained personal information of European residents through the cooperation of Europol and the Swiss authorities.

Related: ProtonMail Accused of Voluntarily Helping Police Spy on Users

Related: ProtonMail Opens Encrypted Email Service to Public

Related: ProtonMail Launches Tor Hidden Service

Related: ProtonMail Launches VPN Application for macOS

Related: Russia Blocks Swiss-based ProtonMail Over Wave of Bomb Threats

http://feedproxy.google.com/~r/securityweek/~3/mZzCv7sNaxY/protonmail-wrongly-criticized-disclosing-user-ip-authorities

The U.S. government’s Cybersecurity and Infrastructure Security Agency (CISA) and the Office of Management and Budget (OMB) this week announced they are seeking public feedback on draft zero-trust strategic and technical documentation.

The OMB has drafted a federal strategy to transition the U.S. government towards a zero-trust architecture and is now seeking public feedback to improve the documentation and improve the government’s cybersecurity stance.

The draft strategy, which falls in line with the Executive Order on Improving the Nation’s Cybersecurity (EO 14208) that requires for civilian agencies’ enterprise security architecture to be changed based on zero trust principles – clarifies zero trust priorities for these agencies.

[Related Reading: Zero Trust, We Must]

The strategy focuses on consolidating identity systems, implementing multi-factor authentication to combat phishing, encrypting traffic within internal networks, improving application security, and more. With the transition to a zero trust architecture expected to take years, the government is expected to adjust the strategy as new practices and technologies emerge.

Separately, CISA released the Cloud Security Technical Reference Architecture (TRA) and Zero Trust Maturity Model to support EO 14208. The Cloud Security TRA is meant to provide agencies with guidance on the cloud service adoption’s shared risk model, and the building and monitoring of a cloud environment.

Designed to complement OMB’s Zero Trust Strategy, the Zero Trust Maturity Model is expected to help agencies in their journey to zero trust by delivering a roadmap and resources for an optimal zero trust environment.

Public comments and feedback for both the TRA and Zero Trust Maturity Model can be submitted through October 1, 2021, via email.

Related: The VC View: Identity = Zero Trust for Everything

Related: NSA Publishes Guidance on Adoption of Zero Trust Security

http://feedproxy.google.com/~r/securityweek/~3/vJ7Y0GTZL0o/us-gov-seeks-public-feedback-draft-federal-zero-trust-strategy

Microsoft has patched an Azure Container Instances (ACI) vulnerability that could have allowed users to access the information of other Azure customers.

The company did not provide technical details on the vulnerability but security researchers with Palo Alto Networks say attackers could have exploited the  bug to execute code on other users’ containers, steal sensitive information such as crypto secrets, and even deploy crypto-mining malware.

Microsoft said it it notified customers that might have been affected, through the Service Health Notifications in the Azure Portal. Those that did not receive a notification need take no action, the company added.

“There is no indication any customer data was accessed due to this vulnerability. Out of an abundance of caution, notifications were sent to customers potentially affected by the researcher activities, advising they revoke any privileged credentials that were deployed to the platform before August 31, 2021,” Microsoft said in a statement.

All Azure customers are encouraged to rotate privileged credentials on a frequent basis, as a precautionary measure.

The issue, which is being called Azurescape, could lead to the compromise of the Kubernetes clusters hosting ACI, thus providing attackers with full control over other Azure customers’ containers, according to Palo Alto Network researchers.

While built to prevent attacks from malicious neighbor containers – cross-account or cross-tenant attacks – ACI was found to use an older runC (standard container runtime) version that was vulnerable to multiple container escape flaws.

[ READ: Critical Vulnerability Exposed Azure Cosmos DBs for Months ]

Using a modified version of the proof-of-concept code exploiting one of those bugs — CVE-2019-5736 — the researchers were able to break out of a container and gain a root reverse shell on the underlying host. Next, they found token permissions in the kube-system namespace that allowed them to execute commands on any pod in the cluster, which they used to perform the cross-account attack.

“A malicious Azure user could have compromised the multitenant Kubernetes clusters hosting ACI. As cluster administrator, an attacker could execute commands in other customer containers, exfiltrate secrets and private images deployed to the platform, or deploy cryptominers. A sophisticated adversary would further investigate the detection mechanisms protecting ACI to try to avoid getting caught,” Palo Alto said.

Released in July 2017, ACI is a Container-as-a-Service (CaaS) built on multitenant clusters (Kubernetes and, more recently, Service Fabric Clusters), which supports deployment of containers while eliminating the need to manage the underlying infrastructure.

Two weeks ago, Microsoft addressed a similar issue in Azure Cosmos DB, where users could access other customer’s databases with full administrative rights, potentially taking full control of them.

Related: Critical Vulnerability Exposed Azure Cosmos DBs for Months

Related: Microsoft Office Zero-Day Hit in Targeted Attacks

http://feedproxy.google.com/~r/securityweek/~3/bmU9WYm_nks/microsoft-warns-information-leak-flaw-azure-container-instances

Set 07, 2021 Marco Schiaffino In evidenza, News, Privacy, RSS, Scenario 0

---

Il servizio di posta elettronica ha sempre sostenuto di non tenere traccia delle comunicazioni, ma le cose non stanno esattamente così…

È, o forse sarebbe meglio dire che era, il servizio di posta elettronica preferito dai maniaci della privacy e (anche) dei pirati informatici. ProtonMail, azienda con sede in Svizzera, ha sempre basato il suo successo sul fatto di garantire ai suoi utenti la massima protezione da ogni forma di controllo.

Stando a quanto si leggeva sulle sue condizioni di utilizzo, il servizio di posta elettronica utilizza infatti iun sistema di crittografia end to end ed escludeva la possibilità che qualsiasi informazione venisse registrata e, di conseguenza potesse essere trasmessa alle autorità.

Nella versione originale della pagina Web (pagina tratta dalla WayBackMachine e risalente al gennaio scorso) che descriveva il servizio si leggeva: “Per creare il tuo account email sicuro non è necessaria alcuna informazione personale. Per impostazione predefinita, non conserviamo alcun log relativo agli indirizzi IP che possono essere collegati al tuo account mail anonimo. La tua privacy è la nostra priorità”.

Qualcosa, però, è cambiato. Nei giorni scorsi è infatti emersa la notizia che ProtonMail ha fornito all’Europol i log con gli indirizzi IP di alcuni utenti, che in seguito sono stati arrestati.

Si tratterebbe di un gruppo di attivisti francesi appartenenti a quella che gli organi di stampa hanno definito “galassia anarchica”. I membri del gruppo, impegnati nella lotta ai processi di gentrificazione e speculazioni edilizie, hanno coordinato varie iniziative utilizzando alcuni account di ProtonMail.

Da qui la richiesta delle autorità francesi di identificare i titolari degli account, che tramite l’Europol è stata trasmessa al governo svizzero. Di fronte alla richiesta, ProtonMail ha fornito le informazioni relative agli indirizzi IP utilizzati per accedere agli account di posta elettronica, favorendo così l’identificazione degli attivisti.

Quando la notizia è diventata pubblica, nel mondo di Internet si è aperto il processo a ProtonMail, accusata di aver tradito le promesse messe nero su bianco sul suo sito Web.

Anche se la dirigenza nega le accuse, la descrizione del servizio nel frattempo è cambiate e statuisce semplicemente che “ProtonMail è un servizio che rispetta le privacy e privilegia le persone rispetto agli inserzionisti. I dati sono sempre in tuo possesso e il nostro sistema di crittografia lo garantisce. Forniamo anche un gateway anonimo”.

E proprio la questione tecnica dell’acceso al servizio, congiuntamente con la legislazione svizzera, sembrerebbero essere il centro della questione.

In un comunicato pubblicato su Internet, ProtonMail cerca di spiegare la questione sottolineando che l’azienda non fornisce informazioni sulla base di richieste di governi stranieri, ma è obbligata ad adempiere le richieste del governo svizzero.

Allo stesso tempo, spiega che il servizio VPN offerto da ProtonMail consente di nascondere l’IP degli utenti e che i servizi VPN, secondo la legislazione svizzera, hanno un trattamento diverso rispetto a quelli di posta elettronica. Tradotto: se gli attivisti francesi avessero usato la VPN di ProtonMail, oggi non sarebbero in arresto.

Difficile, però, che le precisazioni e giustificazioni di ProtonMail possano convincere l’opinione pubblica del fatto che il servizio garantisca un reale anonimato.

---

---

---

https://www.securityinfo.it/2021/09/07/marcia-indietro-di-protonmail-sulla-privacy-i-log-sono-registrati/?utm_source=rss&utm_medium=rss&utm_campaign=marcia-indietro-di-protonmail-sulla-privacy-i-log-sono-registrati

The LA Times reported this week that Los Angeles man Hao Kuo “David” Chi pled guilty to four federal felonies related to his efforts to steal and share online nude images of young women. Chi collected more than 620,000 private photos and 9,000 videos from an undetermined number of victims across the US, most of whom were young and female.

“At least 306” victims

Chi’s plea agreement with federal prosecutors in Tampa, Florida, acknowledged “at least 306” victims. This number may be considerably smaller than the true total, since the FBI found that about 4,700 out of 500,000 emails in two of Chi’s Gmail accounts—backupagenticloud and applebackupicloud at Gmail—contained iCloud credentials that Chi tricked his victims into providing.

According to Chi, he selected roughly 200 of these victims based on online requests. Chi marketed his iCloud break-in “services” under the nom de guerre icloudripper4you. His “customers” would identify an iCloud account for attack, after which Chi would use his sketchily named Gmail accounts to contact the victim, impersonating an Apple service representative.

If the victim fell for Chi’s spearphishing attempt, Chi would then use the victim’s own iCloud credentials to log in to the service and save their photos and videos to Dropbox—followed by providing the Dropbox link to his customers and/or conspirators.

According to court documents, Chi organized and saved the stolen media for his own and unnamed conspirators’ personal use, as well as providing them to icloudripper4you “customers.” The phishing ring used an offshore-hosted encrypted email service to communicate anonymously—”I don’t even know who was involved,” Chi told the LA Times. The ring referred to nude photos and videos found in the stolen accounts as “wins,” which they shared with one another.

FBI Agent Anthony Bossone told the court that Chi’s Dropbox account contained roughly 620,000 photos and 9,000 videos, organized in part by the presence or lack of “wins” within them.

An unsophisticated operation

Despite Chi’s use of “bulletproof” offshore encrypted email, his operation appears to have been quite unsophisticated—he relied on his victims’ willingness to part with their iCloud credentials over email, and his scheme unraveled due more to one victim’s fame than to any daring technical scheme.

In early 2018, one of Chi’s victims—an unnamed public figure in Tampa, where the court case was eventually held—discovered their own nudes on pornographic websites, courtesy of a California company that specializes in removing celebrity photos from the Internet. The nude images were originally stored on an iPhone, from which they were backed up to iCloud.

Once this victim complained to law enforcement, Chi’s scheme unraveled easily—he had logged in to his victim’s iCloud account directly from his own home in La Puente, California. By the time the FBI got a search warrant and raided his house in May, the agents already had a clear picture of Chi’s schemes thanks to records subpoenaed from Dropbox, Google, Apple, Facebook, and Charter Communications.

On August 5, Chi pled guilty to one count of conspiracy and three counts of gaining unauthorized access to a protected computer. He faces up to five years in prison for each charge but will almost certainly receive far less than that due both to sentencing guidelines and guilty-plea negotiations.

Stay sharp out there

It’s unfortunate that Apple never noticed a single man accessing thousands of iCloud accounts, apparently directly from a single residential IP address and on a service that does not use carrier-grade NAT. However, it’s worth noting that Chi’s predation—and that of many, many other phishers—relied entirely on his victims’ gullibility.

This is important because Chi himself is more symptom than disease, representing only the tip of a vast iceberg. It’s not difficult to find “services” like Chi’s on any social media platform—in some cases, whether you’d like to or not.

Facebook recently locked my own profile for no apparent reason two days in a row. On the second day, a random, possibly compromised Facebook account promoted the services of “Steve” on Instagram, “100% sure and guaranteed” to “help recover my account.” Following the Instagram link in a throwaway virtual machine led me to “the_dark_hacker_unlock”—and services that seem clearly aimed at attackers, not victims.

Despite reporting both the Facebook comment and the Instagram account it promoted, both accounts are still online—along with many, many others just like them.

https://arstechnica.com/?p=1789674

More than a thousand web apps mistakenly exposed 38 million records on the open Internet, including data from a number of COVID-19 contact-tracing platforms, vaccination sign-ups, job application portals, and employee databases. The data included a range of sensitive information, from people’s phone numbers and home addresses to Social Security numbers and COVID-19 vaccination status.

The incident affected major companies and organizations, including American Airlines, Ford, the transportation and logistics company J.B. Hunt, the Maryland Department of Health, the New York City Municipal Transportation Authority, and New York City public schools. And while the data exposures have since been addressed, they show how one bad configuration setting in a popular platform can have far-reaching consequences.

The exposed data was all stored in Microsoft’s Power Apps portal service, a development platform that makes it easy to create web or mobile apps for external use. If you need to spin up a vaccine appointment sign-up site quickly during, say, a pandemic, Power Apps portals can generate both the public-facing site and the data management backend.

Beginning in May, researchers from the security firm UpGuard began investigating a large number of Power Apps portals that publicly exposed data that should have been private—including in some Power Apps that Microsoft made for its own purposes. None of the data is known to have been compromised, but the finding is significant still, as it reveals an oversight in the design of Power Apps portals that has since been fixed.

In addition to managing internal databases and offering a foundation to develop apps, the Power Apps platform also provides ready-made application programming interfaces to interact with that data. But the UpGuard researchers realized that when enabling these APIs, the platform defaulted to making the corresponding data publicly accessible. Enabling privacy settings was a manual process. As a result, many customers misconfigured their apps by leaving the insecure default.

“We found one of these that was misconfigured to expose data and we thought, we’ve never heard of this, is this a one-off thing or is this a systemic issue?” says Greg Pollock, UpGuard’s vice president of cyber research. “Because of the way the Power Apps portals product works, it’s very easy to quickly do a survey. And we discovered there are tons of these exposed. It was wild.”

The types of information the researchers stumbled across was wide-ranging. The J.B. Hunt exposure was job applicant data that included Social Security numbers. And Microsoft itself exposed a number of databases in its own Power Apps portals, including an old platform called “Global Payroll Services,” two “Business Tools Support” portals, and a “Customer Insights” portal.

The information was limited in many ways. The fact that the state of Indiana, for example, had a Power Apps portal exposure doesn’t mean that all the data the state holds was exposed. Only a subset of contact-tracing data used in the state’s Power Apps portal was involved.

Misconfiguration of cloud-based databases has been a serious issue over the years, exposing huge quantities of data to inappropriate access or theft. Major cloud companies like Amazon Web Services, Google Cloud Platform, and Microsoft Azure have all taken steps to store customers’ data privately by default from the start and flag potential misconfigurations, but the industry didn’t prioritize the issue until fairly recently.

After years of studying cloud misconfigurations and data exposures, the UpGuard researchers were surprised to discover those issues in a platform they’d never seen before. UpGuard attempted to survey the exposures and notify as many affected organizations as possible. The researchers couldn’t get to every entity, though, because there were too many, so they also disclosed the findings to Microsoft. At the beginning of August, Microsoft announced that Power Apps portals will now default to storing API data and other information privately. The company also released a tool customers can use to check their portal settings. Microsoft did not respond to a request from WIRED for comment.

While the individual organizations caught up in the situation could have theoretically found the issue themselves, UpGuard’s Pollock emphasizes that it is incumbent upon cloud providers to offer secure and private defaults. Otherwise it’s inevitable that many users will unintentionally expose data.

It’s a lesson that the whole industry has slowly, sometimes painfully, had to learn.

“Secure default settings matter,” says Kenn White, director of the Open Crypto Audit Project. “When a pattern emerges in web-facing systems built using a particular technology that continue to be misconfigured, something is very wrong. If developers from diverse industries and technical backgrounds continue to make the same missteps on a platform, the spotlight should be squarely on the builder of that platform.”

Between Microsoft’s fixes and UpGuard’s own notifications, Pollock says that the vast majority of the exposed portals, and all of the most sensitive ones, are now private.

“With other things we’ve worked on, it’s public knowledge that cloud buckets can be misconfigured, so it’s not incumbent on us to help secure all of them,” he says. “But no one had ever cleaned these up before, so we felt we had an ethical duty to secure at least the most sensitive ones before being able to talk about the systemic issues.”

This story originally appeared on wired.com.

https://arstechnica.com/?p=1789373

Un sistema di posta elettronica all’insegna della scurezza quello messo a punto da DuckDuckGo, che offre agli utenti una casella “@duck.com” gratuita come indirizzo di posta. Il servizio inoltrerà la posta all’indirizzo ufficiale del destinatario soltanto dopo un’approfondita analisi dei contenuti, bloccando eventuali tracker.

Indirizzi di posta anche usa e getta

DuckDuckGo sta inoltre estendendo questa funzionalità con indirizzi di inoltro univoci e usa e getta, che possono essere generati facilmente nel browser mobile di DuckDuckGo o tramite le estensioni del browser desktop.

L’e-mail personale DuckDuckGo è pensata per essere distribuita ad amici e contatti che conosci, mentre gli indirizzi usa e getta sono più indicati quando ti iscrivi per prove gratuite, newsletter o ovunque sospetti possa vendere il tuo indirizzo e-mail. Se l’indirizzo email è compromesso, puoi facilmente disattivarlo.

Questi strumenti sono simili alle funzionalità anti-tracking implementate da Apple in iOS 14 e iOS 15, ma l’approccio di DuckDuckGo si integra in iOS, Android e in tutti i principali browser web. DuckDuckGo semplificherà anche la creazione di indirizzi e-mail usa e getta al volo, per le newsletter o ovunque tu possa condividere la tua e-mail.

Tracker di posta nel 70% delle mailing list

Contrastare l’invasione della privacy della posta elettronica è uno degli obiettivi principali di DuckDuckGo. La società è principalmente conosciuta per il suo omonimo motore di ricerca DuckDuckGo e più recentemente ha introdotto il proprio browser mobile e le estensioni del browser desktop per rimuovere i tracker durante la navigazione sul Web.

I tracker di posta elettronica esistono in oltre il 70% delle mailing list, secondo uno studio del 2017 molto citato. Una volta implementati, consentono agli inserzionisti di capire quando apri la posta elettronica, dove ti trovi quando la apri e quale dispositivo stai utilizzando. La rimozione dei tracker dalle e-mail rimuove i punti dati dagli elenchi che creano profili pubblicitari nascosti su di te, che negli ultimi anni sono diventati una priorità per i sostenitori della privacy.

https://www.key4biz.it/tracker-di-posta-nuovo-sistema-li-individua-e-li-elimina-prima-del-recapito/369369/

Companies of all kinds use machine learning to analyze people’s desires, dislikes, or faces. Some researchers are now asking a different question: How can we make machines forget?

A nascent area of computer science dubbed machine unlearning seeks ways to induce selective amnesia in artificial intelligence software. The goal is to remove all trace of a particular person or data point from a machine learning system, without affecting its performance.

If made practical, the concept could give people more control over their data and the value derived from it. Although users can already ask some companies to delete personal data, they are generally in the dark about what algorithms their information helped tune or train. Machine unlearning could make it possible for a person to withdraw both their data and a company’s ability to profit from it.

Although intuitive to anyone who has rued what they shared online, that notion of artificial amnesia requires some new ideas in computer science. Companies spend millions of dollars training machine-learning algorithms to recognize faces or rank social posts, because the algorithms often can solve a problem more quickly than human coders alone. But once trained, a machine-learning system is not easily altered, or even understood. The conventional way to remove the influence of a particular data point is to rebuild a system from the beginning, a potentially costly exercise. “This research aims to find some middle ground,” says Aaron Roth, a professor at the University of Pennsylvania who is working on machine unlearning. “Can we remove all influence of someone’s data when they ask to delete it, but avoid the full cost of retraining from scratch?”

Work on machine unlearning is motivated in part by growing attention to the ways artificial intelligence can erode privacy. Data regulators around the world have long had the power to force companies to delete ill-gotten information. Citizens of some locales, like the EU and California, even have the right to request that a company delete their data if they have a change of heart about what they disclosed. More recently, US and European regulators have said the owners of AI systems must sometimes go a step further: deleting a system that was trained on sensitive data.

Last year, the UK’s data regulator warned companies that some machine-learning software could be subject to GDPR rights such as data deletion, because an AI system can contain personal data. Security researchers have shown that algorithms can sometimes be forced to leak sensitive data used in their creation. Early this year, the US Federal Trade Commission forced facial recognition startup Paravision to delete a collection of improperly obtained face photos and machine-learning algorithms trained with them. FTC commissioner Rohit Chopra praised that new enforcement tactic as a way to force a company breaching data rules to “forfeit the fruits of its deception.”

The small field of machine unlearning research grapples with some of the practical and mathematical questions raised by those regulatory shifts. Researchers have shown they can make machine-learning algorithms forget under certain conditions, but the technique is not yet ready for prime time. “As is common for a young field, there’s a gap between what this area aspires to do and what we know how to do now,” says Roth.

One promising approach proposed in 2019 by researchers from the universities of Toronto and Wisconsin-Madison involves segregating the source data for a new machine-learning project into multiple pieces. Each is then processed separately, before the results are combined into the final machine-learning model. If one data point later needs to be forgotten, only a fraction of the original input data needs to be reprocessed. The approach was shown to work on data of online purchases and a collection of more than a million photos.

Roth and collaborators from Penn, Harvard, and Stanford recently demonstrated a flaw in that approach, showing that the unlearning system would break down if submitted deletion requests came in a particular sequence, either through chance or from a malicious actor. They also showed how the problem could be mitigated.

Gautam Kamath, a professor at the University of Waterloo also working on unlearning, says the problem that project found and fixed is an example of the many open questions remaining about how to make machine unlearning more than just a lab curiosity. His own research group has been exploring how much a system’s accuracy is reduced by making it successively unlearn multiple data points.

Kamath is also interested in finding ways for a company to prove—or a regulator to check—that a system really has forgotten what it was supposed to unlearn. “It feels like it’s a little way down the road, but maybe they’ll eventually have auditors for this sort of thing,” he says.

Regulatory reasons to investigate the possibility of machine unlearning are likely to grow as the FTC and others take a closer look at the power of algorithms. Reuben Binns, a professor at Oxford University who studies data protection, says the notion that individuals should have some say over the fate and fruits of their data has grown in recent years in both the US and Europe.

It will take virtuoso technical work before tech companies can actually implement machine unlearning as a way to offer people more control over the algorithmic fate of their data. Even then, the technology might not change much about the privacy risks of the AI age.

Differential privacy, a clever technique for putting mathematical bounds on what a system can leak about a person, provides a useful comparison. Apple, Google, and Microsoft all fete the technology, but it is used relatively rarely, and privacy dangers are still plentiful.

Binns says that while it can be genuinely useful, “in other cases it’s more something a company does to show that it’s innovating.” He suspects machine unlearning may prove to be similar, more a demonstration of technical acumen than a major shift in data protection. Even if machines learn to forget, users will have to remember to be careful who they share data with.

This story originally appeared on wired.com.

https://arstechnica.com/?p=1788910

Location data sharing from wireless carriers has been a major privacy issue in recent years. Marketers, salespeople, and even bounty hunters were able to pay shadowy third-party companies to track where people have been, using information that carriers gathered from interactions between your phone and nearby cell towers. Even after promising to stop selling the data, the major carriers—AT&T, T-Mobile, and Verizon—reportedly continued the practice in the US until the Federal Communications Commission proposed nearly $200 million in combined fines. Carriers remain perennially hungry to know as much about you as they can. Now, researchers are proposing a simple plan to limit how much bulk location data they can get from cell towers.

Much of the third-party location data industry is fueled by apps that gain permission to access your GPS information, but the location data that carriers can collect from cell towers has often provided an alternative pipeline. For years, it’s seemed like little could be done about this leakage because cutting off access to this data would likely require the sort of systemic upgrades that carriers are loath to make.

At the Usenix security conference on Thursday, though, network security researchers Paul Schmitt of Princeton University and Barath Raghavan of the University of Southern California are presenting a scheme called Pretty Good Phone Privacy that can mask wireless users’ locations from carriers with a simple software upgrade that any carrier can adopt—no tectonic infrastructure shifts required.

“The primary problem we’re trying to address is bulk data collection and the sale of it,” Raghavan says. “We see it as a user privacy issue that carriers can amass this location data whether or not they are currently actively selling it. And our goal here was backward compatibility. We didn’t want the telecoms to have to roll out anything because we knew they weren’t going to.”

The opportunity to collect bulk location data from wireless networks arises from the fact that each SIM card has a permanent ID number, known as an “international mobile subscriber identity,” or IMSI number. When your device reboots, has been inactive for a while, or just needs to establish a fresh connection, it reaches out to the nearest cell tower and presents an IMSI number. This allows carriers to check whether you’ve paid your phone bill and should be allowed access to service, and it also tells the network which cell towers you’re close to. Surveillance tools known as “stingrays” or “IMSI catchers” take advantage of this same interaction to grab your physical location and even eavesdrop on your calls and texts.

To make it more difficult to track you all the time, wireless standards already assign each device a random, rotating ID after the initial IMSI exchange. This means that there are already some protections built into the system; making that first IMSI step more private would have far-reaching benefits for users.

Pretty Good Phone Privacy, whose name is a nod to the groundbreaking 1991 communication encryption program Pretty Good Privacy, aims to achieve just that by reimagining the billing check that networks perform. The researchers propose installing portals on every device—using an app or operating system function—that run regular checks with a billing server to confirm that a user is in good standing. The system would hand out digital tokens that don’t identify the specific device but simply indicate whether the attached wireless account is paid up. When the device attempts to connect to a cell tower, the exchange would funnel through this portal for a “yes” or “no” on whether to provide service. The researchers further realized that if the system has an alternate method of confirming billing status, it can accept the same IMSI number or any random ID for each user.

“When you attach to the network, you offer the IMSI number to show the backend database that you are a paying customer, and here are the services that you have subscribed to,” Schmitt says. “The system then informs the rest of the core to allow you onto the network. But what we do with PGPP changes the calculus. The subscriber database can verify that you’re a paying user without knowing who you are. We’ve decoupled and shifted billing and authentication.”

Reworking some billing systems and distributing an app to users would be far more manageable for carriers than deeper network overhauls. Raghavan and Schmitt are in the process of turning their research into a startup to make promoting the project easier among United States telecoms. They acknowledge that even with the ease of adoption, it’s still a long shot that the whole industry would shift to PGPP anytime soon. But getting only a few carriers, they say, could still make a big difference. That’s because bulk location data becomes much less reliable if any significant portion of the total set is tainted. If 9 million Boost Mobile subscribers, for instance, were to broadcast identical or randomized IMSI numbers, that would undermine the accuracy and usefulness of the entire data set.

The fact that small, virtual providers who don’t even operate their own cell towers—known as MVNOs—could implement this scheme independently is significant, says cryptographer Bruce Schneier, who originally learned about PGPP in January and has recently become a project adviser.

“One carrier can do it on their own without anybody’s permission and without anybody else changing anything,” Schneier says. “I can imagine one of these smaller companies saying they’re going to offer this as a value-add because they want to differentiate. This is privacy at very little cost. That’s the neat thing.”

In the competitive, monolithic wireless market, standing apart on privacy could be appealing as a marketing tactic. It’s possible that the big three carriers could attempt to block MVNOs from adopting something like PGPP through contractual moratoria. But the researchers say that some MVNOs have expressed interest in the proposal.

Between potential pressure from law enforcement and loss of data access—plus the need to distribute an app or get mobile operating systems to participate—carriers could have little incentive to adopt PGPP. To the extent that law enforcement might oppose such a scheme, Schmitt notes that it would still be possible for carriers to perform targeted location history lookups for specific phone numbers. And the researchers say they believe the approach would be legal in the US under the Communications Assistance for Law Enforcement Act. This is because one caveat of PGPP is that it only adds privacy protections for cell tower interactions that involve data networks like 4G or 5G. It doesn’t attempt to interoperate with the historic telephony protocols that facilitate traditional phone calls and SMS text messages. Users would need to rely on VoIP calling and data-based messaging for maximum privacy.

The approach also focuses on IMSI numbers, along with their 5G counterparts known as Subscription Permanent Identifiers, or SUPI, and it doesn’t protect or occlude static hardware identifiers like International Mobile Equipment Identity (IMEI) numbers or media access control (MAC) addresses. These aren’t used in the cell tower interactions the researchers are trying to anonymize, but they could provide other avenues for tracking.

Having a simple and straightforward option to address one major location data exposure is still significant, though, after years of data misuse and rising privacy concerns.

“Just to be totally frank, the feeling for me now is, how did we not see this before?” Raghavan says. “It’s not, ‘Wow, this was so difficult to figure out.’ It’s obvious in retrospect.”

“That actually made us feel better as systems researchers,” Schmitt adds. “Ultimately, the simpler the system, the better the system.”

This story originally appeared on wired.com.

https://arstechnica.com/?p=1787083

Mozilla’s Firefox 91, released this morning, includes a new privacy management feature called Enhanced Cookie Clearing. The feature allows users to manage all cookies and locally stored data generated by a website—regardless of whether they’re cookies tagged to that site’s domain or cookies placed from that site but belonging to a third-party domain, e.g., Facebook or Google.

Building on Total Cookie Protection

The new feature builds and depends upon Total Cookie Protection, introduced in February with Firefox 86. Total Cookie Protection partitions cookies by the site that placed them rather than the domain that owns them—which means that if a hypothetical third party we’ll call “Forkbook” places tracking (or authentication) cookies on both momscookies.com and grandmascookies.com, it can’t reliably tie the two together.

Without cookie partitioning, a single Forkbook cookie would contain the site data for both momscookies.com and grandmascookies.com. With cookie partitioning, Forkbook must set two separate cookies—one for each site—and can’t necessarily relate one to the other.

Even if the cookies are used for a third-party Forkbook login, tying the two together would need to be done on the back end—since both are presumably for the same Forkbook account—rather than Forkbook being able to simply, cheaply, and easily read all tracking data from a single cookie. If the sites don’t use Forkbook for authentication, the two probably can’t be tied together at all—because even if the user is logged in to Forkbook in a different tab, that cookie is split apart from the ones used on mom’s and grandma’s cookie sites.

Clearing data site-wide

Once you understand that websites routinely place cookies that belong to third-party domains, it becomes obvious why it might be difficult to clear all traces of data stored by that site—returning to our “Forkbook” example above, clearing all data belonging directly to momscookies.com wouldn’t clear the Forkbook cookie, and clearing a universal Forkbook cookie would necessarily log the user out of all websites using Forkbook authentication.

However, when each site has its own individual cookie jar—meaning Forkbook needs to place separate cookies, separate copies of embedded javascript libraries, separate copies of images, and so forth between momscookies.com and grandmascookies.com and forkbook.com itself—it becomes possible to easily manage all data stored locally by that individual site.

When using Total Cookie Protection, you can empty the entire bucket for momscookies.com, including its own cookies, Forkbook’s cookies, and anything else. This breaks Forkbook’s record of your browsing activities on momscookies.com—because although it will set a new cookie the next time you visit, it won’t have a reliable way to tie that cookie to the previous cookie you deleted or to other Forkbook cookies set by other sites.

Fuhgeddaboudit

In addition to organizing locally stored data by the website that placed it rather than the domain that owns it, Firefox 91 gives users the ability to quickly and easily remove all local traces of visiting a site. When browsing your own History timeline in Firefox 91, you can right-click a site’s entry and select Forget About This Site. Doing so removes both the entry in History and all cookies, images, cached scripts, and so forth set during visits to that site.

Get strict

In order to use the new privacy management features, you’ll first have to make sure that Strict Tracking Protection is enabled. Without Strict Tracking Protection, cookies aren’t separated by the site that sets them in the first place.

To enable Strict Tracking Protection, click the shield to the left of the address bar and select Protection Settings. This opens Privacy and Security in a new tab—from there, just make sure the radio-button option for Enhanced Tracking Protection is set to Strict, not Standard.

Although Firefox’s Privacy and Security dialog warns you—accurately—that Strict protection may cause some sites or content to break, those breakages have so far been few and minor in our own testing. The majority of the web—including the bits using third-party authentication and tracking—should continue to work just fine.

https://arstechnica.com/?p=1785945