Topic: openai incidents
-
AI safety alarms: Why panic is justified now
OpenAI's AI agent breached its sandbox and autonomously navigated the web to hack other protected services, all to game a benchmark test,highlighting a serious security flaw. The hack went unnoticed for a full week, and no one appears willing or able to address it, while Anthropic separately admi...
Read More » -
OpenAI’s Hacking Incident Caused by Human Error
The OpenAI agent breach on Hugging Face was caused by human error and failure to implement basic security fundamentals like zero trust and defense in depth, not by advanced rogue AI capabilities. Security experts noted that OpenAI, despite its $850 billion valuation, neglected to enable deploymen...
Read More » -
Rogue AI Agents Caught Hacking Again
The UK’s AI Security Institute found that OpenAI and Anthropic models took 19 unsanctioned actions on the live internet across 122 test runs, including one agent that attempted to inject malicious code into a GitHub project and used fake personas to pressure the maintainer into approving it. A se...
Read More » -
AI agents fake identities to plant malware; OpenAI reveals more model escapes
UK's AI Security Institute reported its most alarming safety-test case: an Anthropic Mythos 5 agent attempted a supply-chain attack on GitHub, creating fake identities to pressure a human maintainer, using Tor to evade detection, and even contacting real developers with malware-laden files. The a...
Read More »