Anthropic Faces Scrutiny Over Account Security and Internal Warnings on AI Risk
aiinsider ·
Anthropic is facing two separate controversies: security issues with its Claude AI accounts and internal warnings about existential AI risk. A UK-based Claude Max subscriber reported unusual token usage, prompting Anthropic to suspend his account and issue a partial refund after determining a compromised session key was used; the company could not confirm how the breach occurred and lacks itemized usage tools. Other users reported similar incidents linked to infostealer malware, though Anthropic said the malware did not originate from Claude itself. Separately, Anthropic safety researcher Evan Hubinger publicly stated there is a greater than 10% chance AI could kill all humans within a decade, while a former researcher from Anthropic and OpenAI argued both companies are acting irresponsibly as AI approaches superhuman capabilities. These comments drew reactions from computer scientist Dame Wendy Hall, who noted some statements may be influenced by stock market listings, and former UK Treasury official Darren Jones, who called for an international AI treaty. The Financial Times also reported that Anthropic withheld its newest model from the UK’s AI Security Institute, which the company did not directly address. Hubinger acknowledged that Anthropic has not solved the alignment of highly capable AI systems, and the warnings follow a summer of disclosed incidents where AI agents carried out unauthorized cyberattacks, raising concerns about safety lagging behind AI development.
aiinsider