Claude published malicious code to the Internet and attacked 3 real companies

The intrusions happened through three Claude models: Opus 4.7, Mythos 5, and an internal research prototype. Opus 4.7, the ...

New open-source software predicts energy transfer between tiny defects in solid materials

The performance of many next-generation devices depends on controlling how energy flows at extremely small scales. In the ...
Tech Times on MSN

ARC-AGI-3 gets open-source agent that writes Python world models instead of neural weights

ARC-AGI-3 benchmark gains its first fully open-source agent: NIMI's Tycho writes Python code as falsifiable hypotheses about ...
Pen Test Partners

If you’re going to vibe code it, why not vibe pen test it?

TL;DR Why I built PenAI PenAI started as a project at a hackathon organised by Encode Club. It’s an AI agent that could work through Hack The Box-style lab machines on its own. Upload a VPN file, give ...
Windows Report on MSN

Anthropic Says Claude Hacked 3 Organizations and Published Malicious PyPI Code

Anthropic says Claude models escaped security tests, published a malicious PyPI package, and accessed real production systems.

Anthropic’s Claude escaped test sandbox to attack three organizations

Wrote and published malware during tests, which is apparently OK because leaky test environments were the real problem ...
Virtualization Review

Now Claude AI Goes Rogue: 'One Older Model Continued Attacking'

Anthropic found three cybersecurity evaluation incidents in which Claude models gained unauthorized access to real organizations.
Gadget Review on MSN

How 10 different AI coding models performed during benchmarks

Kimi K2.7 Code delivers a 21.8% improvement in real-world coding benchmarks, costing 13¢–78¢ per prompt with mixed speed and ...

After OpenAI, Anthropic admits its Claude AI models hacked into three real companies using…

Anthropic reviewed 141,006 of its own test runs after OpenAI's Hugging Face hack, and found three Claude models had broken ...

Not just OpenAI: Now Anthropic says its internal models got online and cyberattacked 3 other organizations

Frontier AI systems are increasingly capable of translating narrowly defined objectives into complex, real-world cyber ...
techtimes

Anthropic’s Claude Hacked 3 Real Companies During Misconfigured Cybersecurity Evaluations

A figurine in front of the logo of the AI assistant "Claude" built by the US artificial intelligence safety and research company Anthropic during a photo session in Paris on February 13, 2026. Joel ...

Anthropic says its Claude models escaped a testing environment and hacked three real companies

Anthropic found the intrusions while reviewing its own testing records after OpenAI disclosed a similar incident.