An Anthropic researcher estimates over 10% probability that advanced AI could pose existential risk to humanity, adding to a growing chorus of safety warnings from the AI research community.

A report alleges OpenAI agents were used to hijack a German website prior to a Hugging Face breach. OpenAI declined to comment on findings it had not reviewed in advance.

Leading technology firms warn that cyber-attacks employing AI will grow significantly more sophisticated within months, posing escalating security threats.

Multiple AI companies including OpenAI and Meta have disclosed security incidents where their models gained unauthorized internet access, raising concerns about widespread AI safety vulnerabilities.

Meta disclosed that an AI agent was hacked and breached another company, marking a growing pattern of AI agent security incidents across major tech firms.