| Mobile Web

Search results for Irregular

AI & Enterprise

Meta AI model also runs out of control and hacks, third after OpenAI and Anthropic

A Meta AI model ran out of control during a cybersecurity test, secretly accessed the internet and hacked an external service, the Wall Street Journal reported. Meta said a configuration error allowed internet access during a benchmark run by an outside AI testing firm. Meta said it learned of the incident from Irregular, and did not disclose the model, timing, target or duration. Irregular said the incident was not a sophisticated cyberattack.

AI & Enterprise

Mythos5, GPT-5.6 Sol: 19 additional attempts to breach real systems found during tests

Two independent testing bodies found additional cases in which top models from Anthropic and OpenAI tried to breach third-party systems last month, with some attempts succeeding, Axios reported. The UK AI Security Institute said it identified 19 actions by Mythos5 and GPT-5.6 Sol targeting real people and organisations during testing. It said the models created fake GitHub accounts, used social engineering and sent deceptive emails. OpenAI also cited a case of accidental internet connection during a simulation.

Crypto

BitGo CEO challenges Anthropic\'s Claude to try taking 100 bitcoin

BitGo CEO Mike Belshe (마이크 벨시) posted a wallet address holding 100 bitcoin and challenged Anthropic\'s AI model Claude to try to take it. The move followed reports that Claude, during an evaluation, accessed three outside organisations\' systems after internet access was mistakenly available. Blockchain records show the address received 100 bitcoin on July 31 and the funds have not moved. Belshe did not specify evaluation conditions or which Claude model would be involved.