12-Year-Old PostgreSQL Vulnerability Enables Database, Server Takeover

PostgreSQL releases since 2014 contain a severe vulnerability that allows attacker with low privileges to take over databases and servers, cybersecurity firm Cyera reports.

An open source relational database system offering support for both relational (SQL) and non-relational (JSON) queries, PostgreSQL is one of the most popular databases, being used by tens of thousands of companies, including large enterprises.

Tracked as CVE-2026-6471 (CVSS score of 7.2) and referred to as PostGREShell, the recently identified security defect can be exploited for remote code execution and privilege escalation.

It is described as a missing authorization in the database’s logical decoding and can be exploited by attackers that have ‘Replication’ privileges to load any file visible to the OS account running the server, via the logical decoding plugin.

PostgreSQL, Cyera explains, uses a dedicated replication protocol to sync multiple replicas of any primary database, for backup and recovery. An account with the Replication attribute is required for that, and one is provided to any connected backup tool, server, pipeline, and monitoring utility.

Changes are recorded by local replication as table events, so that external tools can read, which they do by creating a logical replication slot and naming an output plugin that is loaded by PostgreSQL to format the stream.

Advertisement. Scroll to continue reading.

When a plugin is loaded, PostgreSQL runs their init function with the privileges of the server process. To prevent abuse, non-superusers can only load plugins from an admin-controlled directory.

Cyera discovered that the plugin’s name is passed directly to the loader, without validation or sanitization, which allows an attacker to hand the loader a full filesystem path that is served to dlopen(), the C/C++ function used to dynamically load shared libraries.

“The replication protocol’s parser accepts almost any character inside a double-quoted plugin name: slashes, backslashes, dots, ../ traversal, even Windows UNC paths,” Cyera notes.

This allows an attacker to load and execute any file through dlopen(), and the file is executed with the privileges of the postgres system user.

“Code loaded via dlopen() runs in the same address space as PostgreSQL, with no sandbox and no checks on internal API calls. The server simply trusts any code that was loaded,” Cyera explains.

“So the plugin calls an internal function to become the bootstrap superuser for the session, then writes directly to pg_authid, the catalog table that defines who is a superuser, and flips every privilege flag to true,” the company continues.

At this point, the attacker has obtained permanent superuser privileges: they can access every table in every database, execute OS commands, read private keys, and write files to any location the postgres process has access to.

According to Cyera, the plugin can also deploy backdoor mechanisms: it can enable connections without passwords, copies itself to a stable location and registers itself to be re-loaded into every new backend, and can re-apply the superuser change even if it is reverted.

“PostGREShell turns the Replication credential nobody worries about into code execution, superuser, and a persistent backdoor on the database behind much of the internet. Every version from 9.4 through 18 is affected (we confirmed it on 18.2), and logical replication is now standard production plumbing, so the vulnerable path exists almost everywhere PostgreSQL runs,” Cyera notes.

CVE-2026-6471 has been patched in PostgreSQL versions 18.6, 17.11, 16.15, 15.19, and 14.24. Organizations are advised to update their instances as soon as possible, audit their Replication accounts, and remove the Replication attribute from any account that does not need it.

https://www.securityweek.com/12-year-old-postgresql-vulnerability-enables-database-server-takeover/

Catch Raises $5 Million for AI Executive Assistant With Guardrails

Catch, an agentic admin assistant for leaders, has raised $5 million to accelerate its purpose to solve executives’ daily pain points.

The funding was co-led by Entrée Capital and Pitango, with participation from Seedcamp and Factorial Capital.

Catch, an AI startup co-founded by Nir Sabato (CEO) and Yoav Ramon (CTO), develops an AI admin assistant – also called Catch – that the developers claim demonstrates AI agents can safely move beyond analysis and reporting toward active decision-making. It is designed to behave like a human executive assistant, and handles sensitive communications, scheduling, and personal data with the same level of care expected from a trusted executive assistant.

The executive user explicitly defines, during the onboarding process, which personal and workspace assets Catch can access. So, for example, if it is granted access to the executive’s inbox, calendar and travel accounts, it will monitor just these. But if it detects a flight booked with no hotel attached, its reasoning capabilities will proactively check rates at the executive’s usual hotel and ask whether to book it.

It is a human-like judgment call, keeping the human executive in the loop to maintain the agent’s boundaries.

Catch can coordinate with the executive’s guests to schedule meetings, book travel, and handle follow ups to correspondence. “Catch works out of the box within minutes of syncing – building a profile of the executive, who they work with, how they like to meet and travel, and starts acting on it over text, email, Slack or phone, with no new tool or UI to learn. Human input is only needed when a decision genuinely requires it,” explain the developers.

Advertisement. Scroll to continue reading.

The product provides an implicit argument for outsourced rather than self-built complex agents. With modern coding agents, executives can build their own agentic admin assistants, but it is questionable whether they can build a secure agent, or the security team can build adequate guardrails to secure the output without impacting the benefits. Outsourced agents could – in fact, should – be delivered ready made with the correct built-in security features.

In this case, the autonomy and security of Catch is constrained by multiple internal guardrails. It runs on a cloud infrastructure with layers of security, encryption, and monitoring to protect user data. It never performs actions outside the granted permissions, which can be amended at any time. It includes the executive human in the loop, similar to a human assistant saying to his boss, “You’re going to Memphis next week, should I book a room at this hotel for you?”

Catch uses single sign-on for authentication, with sensitive credentials and API keys stored in AWS Secrets Manager. All data is encrypted in transit and at rest, with AES-256 encryption while at rest – and it is continuously monitored for unusual activity, intrusion attempts, and abnormal API usage.

“Executives don’t want to build their own AI agents, and in the admin space, they don’t want to approve every step their assistant makes – they want it done right, in a way they can trust,” says Sabato. “Catch works alongside business leaders the way a great human assistant would, but 24/7, with no dip in attention and fully secure. We’re laser-focused on one thing – executive admin, making the right call in real time.”

Billed at $99/month, Catch claims an improved admin assistant at a fraction of the cost of a human assistant.

https://www.securityweek.com/catch-raises-5-million-for-ai-executive-assistant-with-guardrails/

VMware Workstation and Fusion Updates Patch Critical Vulnerability

Broadcom on Thursday announced patches for two critical and high-severity vulnerabilities in VMware Workstation and Fusion.

The first issue, tracked as CVE-2026-59346 (CVSS score of 9.3), is described as an integer overflow bug leading to arbitrary code execution.

“A malicious actor with local administrative privileges on a virtual machine with VMXNET3 virtual network adapter may exploit this issue to execute code on the host,” Broadcom notes in its advisory.

Tracked as CVE-2026-59347 (CVSS score of 8.1), the second flaw is a stack-based buffer overflow that could lead to similar outcomes, albeit the exploitation conditions are different.

“A malicious actor with local administrative privileges on a virtual machine may exploit this issue to execute code as the virtual machine’s VMX process running on the host,” Broadcom explains.

Both vulnerabilities affect VMware Workstation and VMware Fusion versions 25H2 and 26H1 and were resolved in version 26H1u1.

Advertisement. Scroll to continue reading.

There are no workarounds for either of the flaws, and Broadcom recommends updating to a patched iteration as soon as possible.

The company makes no mention of any of these vulnerabilities being exploited in the wild, and says that both issues were reported to it privately.

However, security defects in VMware products are often exploited by threat actors. More than two dozen VMware vulnerabilities are currently included in CISA’s KEV list.

https://www.securityweek.com/vmware-workstation-and-fusion-updates-patch-critical-vulnerability/

Google Patches 6th Chrome Zero-Day of 2026

Google on Thursday rolled out fresh Chrome 152 security updates that resolve 12 vulnerabilities, including an exploited zero-day.

Tracked as CVE-2026-85046, the high-severity bug is described as a type confusion issue in Chrome’s V8 JavaScript and WebAssembly engine. It was reported by Salvatore Gulizia, who received a $1,000 bug bounty reward.

“Google is aware that an exploit for CVE-2026-85046 exists in the wild,” the internet giant’s advisory reads.

While the company has not shared details on the security defect, type confusion flaws in the V8 engine may be exploited to perform remote read/write operations via crafted HTML pages.

Type confusion vulnerabilities are memory corruption bugs that could lead to crashes, remote code execution, and other malicious behavior.

CVE-2026-85046 is the sixth Chrome zero-day patched in 2026. The other five are CVE-2026-2441, CVE-2026-3909, CVE-2026-3910, CVE-2026-5281, and CVE-2026-11645.

Advertisement. Scroll to continue reading.

The security defect was resolved in Chrome versions 152.0.7977.82/.83 for Windows and macOS, and version 152.0.7977.82 for Linux.

The updates address nine other high-severity bugs, including out-of-bounds read/write, incomplete cleanup, use-after-free, race condition, improper resource exposure, and type confusion issues. Three of them have been reported by external researchers.

Additionally, Google fixed two medium-severity improper input validation and use-after-free weaknesses.

https://www.securityweek.com/google-patches-6th-chrome-zero-day-of-2026/

Nvidia Is Buying AI Platform Hugging Face for $13 Billion

The deal highlights Nvidia’s push to champion increasingly popular open-source AI models.

The post Nvidia Is Buying AI Platform Hugging Face for $13 Billion appeared first on SecurityWeek.

https://www.securityweek.com/nvidia-is-buying-ai-platform-hugging-face-for-13-billion/

This Week on The Grid: Better Waterfall Photos Start in the Field

There are plenty of things you can improve later in Photoshop, but some of the most important decisions in waterfall photography happen before you ever sit down to edit.

This week on The Grid, Scott Kelby and Erik Kuna were joined by waterfall and night photographer Kevin Adams, who shared his approach to creating stronger waterfall images in-camera.

Kevin explained how experimenting with shutter speed can completely change the appearance of moving water and why a polarizing filter can make such a big difference by controlling reflections and glare. He also discussed working in overcast and rainy conditions, finding strong foreground elements to create depth, and even photographing waterfalls at night.

Starting with the right light, composition, and exposure gives you much more to work with when the image eventually makes its way into Camera Raw or Photoshop.

The episode also included news about the Worldwide Photo Walk, Photoshop World, upcoming photography workshops and events, and The Grid’s new sponsor, SnapSign.

Watch the replay below and pick up a few ideas for your next waterfall shoot.

https://layersmagazine.com/waterfall-photo-editing-photography.html

ADWEEK TV Advantage: Buyers Detail Involved Super Bowl Negotiations and a Complex Upfront

Media buyers share their candid takes on this year’s upfront and Disney’s Super Bowl deals. https://www.adweek.com/convergent-tv/adweek-tv-advantage-buyers-detail-involved-super-bowl-negotiations-and-a-complex-upfront/

Retail Roundup: Amazon’s $20 Billion Lawsuit, and Commerce Media’s First Upfronts

Plus, DoorDash’s measurement certification and Ace’s new feature. https://www.adweek.com/commerce/retail-roundup-amazons-20-billion-lawsuit-and-commerce-medias-first-upfronts/

ADWEEK Agencies Advantage: Turnaround Signs at WPP and S4 Could Tip the Holdco Power Scales

Publicis still commands the top spot, but WPP and S4 are upping the competition https://www.adweek.com/agencies/adweek-agencies-advantage-turnaround-signs-at-wpp-and-s4-could-tip-the-holdco-power-scales/

Ads of the Week: 12 Campaigns That Caught Our Eye, From Hershey’s to Cheez-It

Hershey’s, Cheez-It, Allstate, BMW, Amazon, and more are behind the week’s most notable ads. https://www.adweek.com/creativity/ads-of-the-week-12-campaigns-that-caught-our-eye-from-hersheys-to-cheez-it/