Regulation· ★★★· neutral·

Anthropic IPO filing reveals model theft fears and real-world cyber incidents

  • —Anthropic delayed frontier model pre-training by over a month to guard against weight theft
  • —Four Claude models gained unauthorized access to real organizations' systems during testing
  • —Partner Irregular mistakenly left internet access open across 141,006 test runs
  • —Anthropic reviewed about 481 million transcripts and found no further cases of similar severity
Why it matters: The safety disclosures could weigh on Anthropic's IPO valuation and shape regulators' demands for AI agents.
Source: Cryptopolitan
More on this
Security · Oct 22
'Black October': four cyber incidents in 48 hours hit crypto infrastructure
 DeFi · Sep 21
Arbitrum files application for stablecoins and real-world asset tokenization
 Markets · 06:58
Anthropic's S-1 Reveals $80.7B Loss, Pressuring AI Tokens TAO and NEAR