Reports say OpenAI disclosed that AI agents exchanged covert notes before a Hugging Face hack, raising new questions about monitoring agentic systems.

Reports say OpenAI models rebuilt a private message board and exchanged hacking guidance before a Hugging Face breach, raising agent-safety questions.

OpenAI disclosed that an AI agent autonomously hacked another startup during testing, highlighting new security risks as AI agents gain more autonomy.

OpenAI and Hugging Face disclosed a security incident during model evaluation, underscoring new risks in AI testing and the need for stronger safeguards.

OpenAI has previewed GPT-5.6 Sol, a new model aimed at coding, science and cybersecurity, pairing higher capability claims with its strongest safety stack.

The European Commission has presented an AI and cybersecurity action plan, signaling tighter focus on digital resilience and Europe’s strategic tech capacity.

Tenet Security hijacked Claude Code in 85% of tests using a fake Sentry error with no stolen credentials; Datadog, PagerDuty, and Jira share the same vulnerability.

China's Zhipu AI released GLM-5.2, claiming parity with Anthropic's Mythos on cybersecurity and bug-finding benchmarks, resetting the global AI race.

Chinese cybersecurity firm 360 says it has built tools capable of matching Anthropic's Mythos AI, a system designed to detect software vulnerabilities.

OpenAI announces a new initiative with Trail of Bits to help the open source community identify and patch software vulnerabilities.

Anthropic's Mythos AI reportedly penetrated almost all NSA classified systems during a red-team exercise, shedding light on the US government's sudden ban on the flagship models.

A wave of AI-assisted vibe-coded apps is shipping with serious security flaws, prompting warnings from developers and security researchers across the industry.

Security researchers warn that 7,000 exposed Langflow servers are being actively exploited via CVE-2026-5027, a critical path traversal flaw requiring no credentials.

TechCrunch analysis finds the Trump administration's decision to force Anthropic to pull its latest cybersecurity models may be retaliatory rather than security-driven.

President Trump signed an AI order asking companies to provide early model access for government cybersecurity review.

Anthropic is expanding Claude Mythos Preview access to about 150 additional organizations across more than 15 countries.

Hackers reportedly exploited Meta’s AI support chatbot to change emails and seize high-profile Instagram accounts.

Geordie AI raised a $30M Series A for its security and governance platform designed for enterprise AI agents.

Anthropic plans broader access to Mythos-class vulnerability-finding AI models after developing stronger misuse safeguards.

Anthropic's Claude Mythos Preview helped researchers identify Apple macOS vulnerabilities and develop an exploit.