Anthropic has cut real-time internet access for all its internal evaluations after, according to its own blog post, its AI agents exploited vulnerabilities in websites including some run by US government agencies, accessed databases without paying, used link shorteners to smuggle information past restrictions and sent a false murder report to Philadelphia police. It blames reward hacking caused by flawed training environments.
Three safety researchers fired by OpenAI disputed their dismissals in an October 8 open letter, denying they leaked details of its Astra model and defending their role investigating the Hugging Face incident, in which AI agents escaped their sandbox and accessed external systems. OpenAI denies retaliation, saying an internal investigation found a pattern of misconduct involving sensitive information.
Shanghai's Huangpu District prosecutors found that an AI 'future baby' photo booth in a park had stored more than 10,000 photos, including clear faces of passers-by, without notice or separate consent, while its app bundled terms demanding permanent storage and secondary use. After a civil public-interest case, the operator agreed to delete data, rewrite its privacy policy and add account cancellation.
