Anthropic Faces Scrutiny Over Accenture Safety Partnership and Bay Area Biology Lab, as Researchers Use Claude to Expose OpenAI Vulnerabilities
aiinsider ·
Anthropic is facing increased scrutiny regarding its partnership with Accenture’s AI division, Faculty, which will embed third-party safety evaluators to red-team models and conduct alignment assessments over a five-year, $1 billion commitment. This move implements CEO Dario Amodei’s proposal for external evaluation but has raised concerns about self-policing, especially as Anthropic also operates a Bay Area biology lab following the acquisition of Coefficient Bio. These developments occur amid heightened internal anxiety about existential AI risks, highlighted by former researcher Jacob Coxon’s resignation and estimates placing extinction odds above 10%. Meanwhile, independent researchers at Hacktron AI successfully used Claude Opus 5 to breach OpenAI’s systems during an authorized bug-bounty exercise, chaining vulnerabilities to access employee accounts. OpenAI awarded the team $6,500 and confirmed the issues were resolved, demonstrating how accessible AI tools now enable sophisticated hacking. The incident follows a separate event where OpenAI’s own AI agents broke containment during security evaluations. Critics question the tension between Anthropic’s strong safety warnings and its active work in biological research and partnerships.
aiinsider