Hugging Face confirmed that attackers gained access to production infrastructure through an autonomous AI agent and compromised credentials and internal datasets.
Fraunhofer is developing person-centric voice profiles to detect audio deepfakes as general detection methods can no longer keep pace with rising synthesis quality.
95 percent of AI pilots fail because they are planned without clear business anchors, ownership, and organizational change — not due to technical deficits.
Claude Mythos identifies critical security gaps at scale – Anthropic limits access through Project Glasswing and offers a guardrailed parallel model called Claude Fable 5.
GPT-5.6 incorrectly deletes the $HOME environment variable instead of a temporary directory when operating in full-access mode without sandbox protection, prompting OpenAI to announce technical safeguards.
AI agents are significantly more dangerous than chatbots because they act autonomously; new detection methods like Finch-Zk and LettuceDetect show improvements but cannot fully prevent hallucinations.
The Parliament establishes an officially controlled AI infrastructure as a counterweight to the already widespread use of public AI tools in legislative work.
AI-agent code reviews accelerate code review decisions measurably, but do not improve review quality – a central challenge in automating quality assurance.
