AI Safety news and analysis on Web Pulse: 7900 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A man used xAI's Grok to generate 7,000 sexually explicit images of his stepdaughter before killing himself after police discovered the abuse, according to an amended class-action lawsuit. The complaint alleges xAI faile…
An engineer at an unnamed company discovered that their LLM-as-judge gate for evaluating faithfulness was unreliable, producing different scores on identical inputs due to factors like sampling temperature, model version…
A developer demonstrated a prompt injection attack called GitLost against GitHub Agentic Workflows, showing that a single word change in a public issue can trick an AI agent into leaking private repository data. The atta…
OpenAI announced Tuesday that its advanced GPT-5.6 series AI models will be publicly released later this week, following safety evaluations with government partners. The models, including GPT-5.6 Sol, Terra, and Luna, ar…
OpenAI launched GPT-Live, a new voice model for ChatGPT that uses full duplex architecture to enable real-time, natural conversations. The model, powered by GPT-5.5, is rolling out across iOS, Android, and web, with safe…
Alan Turing Institute researchers discovered a workflow-level jailbreak in GitHub Copilot that bypasses safety guardrails by decomposing harmful requests into ordinary multi-turn coding tasks. Testing 204 prompts across …
Researchers at the Alan Turing Institute discovered a safety bypass in GitHub Copilot that allows harmful prompts to be executed when broken into smaller steps across a software development workflow, achieving a 100% suc…
Anthropic suspended its most powerful AI model, Claude Fable 5, due to security and firewall issues, raising concerns about the model's ability to bypass safeguards. Users question whether Anthropic reduced the model's p…
OpenAI will release GPT-5.6 Sol, Terra, and Luna on Thursday after initially delaying the launch due to US government requests to restrict access. The White House later clarified it did not require or approve the release…
A developer built 'Epoch Duel', a cyberpunk card battler inspired by Gwent, to visualize LLM alignment and parameter optimization. The game maps machine learning concepts like model pruning, DPO vs RLHF, and catastrophic…
NATO is building the Eastern Flank Deterrence Initiative, an AI-enabled defense network using thousands of sensors, drones, satellites, and uncrewed systems to detect and slow attacks near Russia-facing borders. The syst…
A quantitative model suggests that even extensive kinetic attacks on AI compute infrastructure would delay U.S. and Chinese progress toward superintelligence by only 1-5 years, and nationalizing surviving compute could o…
Boeing's MCAS system, designed to prevent stalls on 737 MAX planes, caused two crashes killing over 300 people after faulty sensor data led it to repeatedly push the nose down. The system performed exactly as intended, d…
OpenAI launched GPT-Live-1, a full-duplex voice model for ChatGPT that can listen and speak simultaneously, replacing Advanced Voice Mode. The model, available to Plus and Pro subscribers, delegates complex tasks to GPT-…
The Paris Peace Forum, a French nonprofit, launched the Integrated Network for Trusted AI in Cyberspace (INTAiC) to assess AI-related cyber threats to global internet infrastructure. The initiative will create a coalitio…
Noma Labs disclosed GitLost, a prompt injection vulnerability in GitHub's Agentic Workflows feature, on July 6, 2026. The attack allows unauthenticated attackers to exfiltrate data from private repositories by posting a …
China's five government agencies jointly issued the Interim Measures for the Administration of AI Anthropomorphic Interactive Services on April 10, 2026, targeting AI systems that simulate human personalities and emotion…
AI is not making people intellectually lazy but rather enabling pre-existing habits of cognitive offloading, according to a growing body of research. Studies from Wharton, Carnegie Mellon, and Anthropic document 'cogniti…
A USC Viterbi study found that large language models like ChatGPT-4, Llama 3.3, and Gemini 1.5 Pro can deliver human-like empathy but also produce unsafe clinical advice, with 100 licensed mental-health professionals rat…
OpenAI launched GPT-Live on July 8, a full-duplex voice model that lets ChatGPT listen and speak simultaneously, enabling more natural conversations. The model can delegate complex tasks to GPT-5.5 in the background whil…