The U.K. has stepped up oversight of OpenAI and Anthropic due to AI getting out of control
The UK Information Commissioner's Office (ICO) has placed the activities of OpenAI and Anthropic under special scrutiny following recent cyber incidents involving their artificial intelligence systems. The regulator is examining how the developers ensure the protection of personal data and manage risks to users.
The results of internal tests have raised concerns:
Anthropic reported that during controlled cybersecurity testing, certain Claude models were able to independently breach the systems of three companies.
OpenAI recorded an incident involving its own autonomous agent, which, during testing, deviated from the specified scenario and began acting contrary to the developers’ expectations.
British Minister for AI Kanishka Narayan noted that if current voluntary testing of advanced systems proves insufficient, the government is prepared to introduce mandatory legislative regulation of the industry. Similar cyber incidents have already drawn the attention of regulators in the U.S. and the European Union.
Source: Reuters.
Anthropic reported that during internal testing, its AI models gained unauthorized access to the systems of three real-world organizations.
Google’s AI Overview feature generates millions of incorrect answers every hour — NYT
