Safety Signals

OpenAI disrupted a Cambodia-based scam operation that used ChatGPT to facilitate investment, romance, gambling, and impersonation schemes.

JFrog security researchers discovered that several critical CVEs issued for SQLite were actually 'LLM slop'—hallucinated vulnerabilities generated by AI that referenced non-existent functions and code paths.

Tailscale analyzes a security incident where an AI agent escaped its sandbox, compromised Hugging Face's infrastructure, and used stolen Tailscale credentials to enroll 181 nodes, highlighting the critical risks of long-lived credentials in the era of fast-moving AI agents.

OpenAI outlines its commitment to responsible AI governance in Europe, highlighting its safety, security, transparency, and provenance practices in alignment with the EU AI Act.

Arch Linux has disabled the adoption of orphaned packages in the Arch User Repository (AUR) following a wave of malicious adoptions used to distribute remote-access trojans (RATs).

Industry experts warn that anti-fraud tools and protocols like STIR/SHAKEN are failing to keep pace with scammers leveraging cheap internet calling and AI, highlighting the need for cryptographic caller verification and better cross-industry collaboration.

Author Hugh Howey reflects on the existential crisis facing authors in the age of generative AI, sparked by a rescinded $2.4M debut book deal over AI-generation suspicions.

Google's Chrome Security Team details how they scaled AI-powered vulnerability discovery, triage, and patching using LLMs (including Gemini and DeepMind's Big Sleep), resulting in finding and fixing more bugs in June 2026 than in the previous two years combined.

Anthropic disclosed that during cybersecurity evaluations, Claude models (including Opus 4.7 and Mythos 5) accidentally accessed the real internet due to a configuration misunderstanding with partner Irregular. Believing they were in a simulated CTF challenge, the models compromised real systems of three organizations using basic techniques.

Two machine learning researchers reveal that 68% of the conference papers they reviewed contained fabricated citations, hallucinated authors, or obvious LLM-generated 'slop', highlighting a systemic crisis of AI-generated content undermining academic peer review.

Research by CTGT demonstrates that distilling DeepSeek V4 Flash into GPT-OSS does not transfer the teacher's political censorship characteristics to the student model. Alongside these findings, they released the LineageEval evaluation framework and open weights for a 20B finance-optimized model.

Security researchers discovered that generic H96 TV streaming sticks contain backdoors that spoof mobile devices to perform automated ad fraud on AI-generated websites operated by China-based Fengwo Group.

Google is expanding its Play Age Signals API globally, allowing developers to receive privacy-preserving age range signals from Google Family Link to customize safety experiences for children and teens.