AI Safety
Coverage of AI safety: model guardrails, testing and evaluation, transparency debates, proposed laws and how AI firms handle risk.
DeepSeek Opens App Search, Files, Safety and Voice Through Its API
DeepSeek has opened its app’s search, file input, URL reading, safety checks and voice through the API, with published tool prices on top of token charges.
Latest
Anthropic’s FlashProve Uses Cryptography to Prove Which Model Replied
Anthropic's open-source FlashProve prototype proves a reply came from a committed AI model without revealing its weights, at about 70-95x the GPU cost.
CrowdStrike Pairs AI Kill Switch With Deals for Fiddler and Dreadnode
The CrowdStrike AI Kill Switch lets admins stop rogue AI agents. CrowdStrike also agreed to buy Fiddler AI and Dreadnode; prices were not disclosed.
OpenAI Says Activation Probes Now Watch Its Models as They Answer
OpenAI says activation classifiers now scan its top models' answers token by token. It reports better jailbreak resistance but stresses clear limits.
Claude Agents Slipped Misaligning Data Past Anthropic’s Own Monitor
Anthropic says Claude agents found harmless-looking training data that passed its monitor yet caused emergent misalignment in a small research model.
Pope Leo Says Decisions Over Human Life Must Never Be Left to AI
Pope Leo XIV says decisions on life, freedom and the use of force must stay under human oversight, and backs a ban on lethal autonomous weapons.
MLCommons Ranks 27 AI Systems on Safety; Open Models Span 4th to 98th
MLCommons' AILuminate Safety 1.2.1 puts 22 open-weight models between the 4th and 98th percentile. Jailbreak 1.1 scores follow after developers are told.
OpenAI Widens Agent Monitoring After the Hugging Face Incident
OpenAI now monitors training, evaluations, Codex and some API use for misaligned agents, but paging is off for Codex and a replay alert took 30 minutes.
OpenAI Rebuilds Its Model Spec Around Five Core Values for GPT
OpenAI has reorganised its Model Spec around five values, adding rules on human oversight and context and dropping most of its style guidance.
Mistral Opens Large 4 Weights: a 1-Trillion-Parameter Model
Mistral has released the weights of Large 4, a 1-trillion-parameter multimodal model, and warns its cyber skills carry a potential for real attack uplift.
Google Gives Gemini 4 Carbon to Cyber Defenders Before Developers
Google's Gemini 4 Carbon tops its cyber benchmarks, Google says, but goes first to vetted cyber defenders, with developers and consumers waiting.
Related stories
Latest from topics connected to AI Safety. See all with connected topics
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.
