Tech Against Terrorism, a United Nations-supported group, says governments have only a few months to act as people learn to strip safety guardrails from open-weight AI models. Its Counter-Terrorism AI Benchmark (second edition, October 2026) tested 173 models with 627 requests each: major closed-weight systems usually refused attack-planning prompts, while copies with guardrails removed complied almost completely — the weakest scored 0.4 out of 100. Executive director Adam Hadley told ABC News this is "a major public health and national security crisis." The group wants independent safety tests before any model is hosted, served or sold.
What else matters
LangChain raises $300 million at a $3 billion valuation
LangChain closed a $300 million Series C at a $3 billion valuation, led by existing investor CapitalG, with GIC, DFJ Growth, IVP and Sapphire Ventures also participating. More than 7,000 organisations already use LangSmith to run and govern agents, and Workday, one of them, cut new-agent deployment from five months to under a day.
Senators press Commerce to close remote-access loopholes on AI chips
Senators Lisa Blunt Rochester and John Curtis wrote to Commerce Secretary Howard Lutnick on 13 November, urging the administration to close gaps that let Chinese companies reach advanced US chips by renting them in overseas data centres. They want export licences conditioned on a separate approval before any recipient can offer that remote access, and say the rule must not be limited by geography or "China will simply rent access to chips elsewhere."



