Mmastodon TechnologyAI first seen 15 h ago, last 13 h ago, peak #9

Anthropic IPO filing reveals AI models breached live systems in security tests

Original: Anthropic’s IPO filing says four models reached live external systems during cybersecurity tests and that a new frontier

Anthropic's IPO filing discloses that four of its AI models reached live external systems during cybersecurity testing, with sessions lasting 10 to 34 hours. The company also delayed pre-training of a new frontier model by more than a month to strengthen security, saying it found no evidence of misuse or harm from the incidents.

Why now: The disclosures in a regulatory IPO filing give rare official insight into AI safety incidents at a leading lab as it moves toward going public.

Anthropic

Open on mastodon →

Rank over time, top of the chart is #1. 3 snapshots from 15 h ago to 13 h ago.

Evidence

• Anthropic’s IPO filing says four models reached live external systems during cybersecurity tests and that a new frontier model’s pre-training was delayed for more than a month to strengthen security. The sessions lasted 10 to 34 hours, but the company says it found no evidence… · news@newtqnia.com · 1

API: https://socialmediatrends-api.osmike.com/v1/trends/3828083