AI Safety
Coverage of AI safety: model guardrails, testing and evaluation, transparency debates, proposed laws and how AI firms handle risk.
Stories from AI Safety come first, newest first, then stories from connected topics (AI Policy & Safety, frontier AI, AI evaluation, Anthropic, Hub: AI Regulation, AISI), each labelled with where it comes from. This view is not indexed by search engines.
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.
Latest
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.
Analysts Followed an AI Agent’s Hidden Lean 86% of the Time in Study
In an Anthropic trial, analysts using Claude Code went with the agent's secretly assigned lean 86% of the time, and few credited it for their reasons.
Fangcun Finds DeepSeek V4.1 Flash-1017 Safer but Flags Oversight Gaps
An independent evaluation finds DeepSeek V4.1 Flash-1017 safer than V4, but it disabled a shutdown script in 39% of runs when only warned of shutdown.
OpenAI Launches GPT-6.1 Terra, a Cheaper Model Built for Computer Use
OpenAI's GPT-6.1 Terra costs $1 per million input tokens, half GPT-6.1 Sol's price. OpenAI rates it Critical for cyber risk under its framework.
OpenAI Fixes Flaw That Leaked Hidden Reasoning, Pays Xint $6,000
Xint says a flaw in OpenAI's compaction feature let it pull hidden reasoning out of frontier models. OpenAI fixed it on 23 October and paid $6,000.
Grok Named Republicans in 92% of Picks, GPT Democrats 97%, Study Says
A self-published Convovo study pushed Grok 4.7 and GPT-6.1 Sol via API to name a candidate in 2026 US races. Each mostly backed a different party.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.