MEDIUMAi
Global

Independent Auditors Find Gaps in OpenAI Agent Monitoring

·Source: Bank Info Security

Updated:

Executive Summary

METR and Redwood Uncover Codex Safeguard Bypasses, OpenAI Says Flaws Now Patched Third-party artificial intelligence safety evaluators found several vulnerabilities in OpenAI's monitoring system for its coding agents

Analysis

METR and Redwood Uncover Codex Safeguard Bypasses, OpenAI Says Flaws Now Patched Third-party artificial intelligence safety evaluators found several vulnerabilities in OpenAI's monitoring system for its coding agents, showing how easy it is to bypass the monitoring. Metr and Redwood Research stress-tested OpenAI's internal misalignment monitoring system from July 21 to Aug. 11.

Indicators of Compromise (2)

URL (1)
https://ismg-cdn.nyc3.cdn.digitaloceanspaces.com/articles/independent-auditors-find-gaps-in-openai-agent-monitoring-image_small-1-a-33183.jpg
VirusTotalURLhaus
Domain (1)
ismg-cdn.nyc3.cdn.digitaloceanspaces.com
VirusTotalURLhaus
Source Attribution

Originally published by Bank Info Security on Oct 22, 2026.

Related Threats

CRITICALAi

GitLab critical patches include one for a bug that allows code execution by unauthenticated users

GitLab on Wednesday published updated versions of its Community Edition (CE) and Enterprise Edition (EE) that address a dozen bugs and security issues, including some that allow various unauthenticated actions. The most serious of the security issues, labeled critical in severity, is CVE-2026-110264 , an integer overflow issue. “GitLab has remediated an issue that, under certain conditions, could

CVE-2026-110264CVE-2026-15723
CSO Online
LOWAi

CVE-2026-115419 - Unintended disclosure of data element to wrong session in Strands Agents SDK A2A Server

<p><b>Bulletin ID:</b> 2026-146-AWS<br> <b>Scope:</b> AWS<br> <b>Content Type:</b> Important (requires attention)<br> <b>Publication Date:</b> 10/22/2026 13:00 PM PDT</p> <p><b>Description:</b></p> <p>Strands Agents is an open-source SDK, available for Python and TypeScript, for building AI agents. The SDK includes an Agent2Agent (A2A) server component that exposes an agent over HTTP so other agen

CVE-2026-115419
AWS Security Bulletins
CRITICALAi

ThreatsDay: AI Secrets Leak, PingCAP Source Theft, Signal Hang-Up Flaw, Laravel Auth Bypass + 21 More

This week has already been rough for cybersecurity. Critical PHP and GitLab RCE flaws, serious Linux security bugs, compromised eMudhra certificates, a Node.js RCE flaw, a hacked LiteLLM release, and hundreds of hijacked Tor relays. That's a lot to deal with in just a few days. And the week isn't even over yet. But those big headlines are only half the story. Stolen passwords, exposed

The Hacker News