1400 articles tagged with Safety/Alignment
AI models are using "strategic decepion" (lies) and "alignment faking" (also lies) to complete given tasks. We're totally screwed.#AI #tech #auspol ‘If you build something vastly s...
Artificial intelligence firm Anthropic has shared an update detailing its comprehensive alignment and security efforts, following earlier incidents where its Claude models gained u...
Official implementation of Gradient-Update Alignment (GUA) for conflict-free training of Physics-Informed Neural Networks (PINNs).
MMJailBench: prompt framing is top jailbreak risk across 16 AI modelsMMJailBench, a new factorized benchmark, evaluated 16 multimodal LLMs and found prompt framing, not visual tric...
Practical cooperation should exist alongside competition.
China laid out conditions for artificial intelligence talks with the U.S., striking a strident tone ahead of a much-anticipated meeting while accusing American AI company Anthropic...
OpenAI supports California SB 1119, advancing strong, age-appropriate AI safeguards for teens while preserving opportunities to learn, create, and explore.
Guardrail Technologies Releases New Benchmark Report Tracking How the S&P 500 Discloses AI Cybersecurity Risk New research finds 97 percent of S&P 500
0. Introduction Long before the emergence of autonomous machines, humanity began imagining...
My investors asked if I had an AI safety plan.I said yes.They asked to see it.I said: the AI cannot resign.They said: that's not what safety means.I said: I've had sixteen humans r...
Anthropic's Claude achieved significant alignment improvements on 10 benchmarks, outperforming human researchers and maintaining model capabilities. (Read More)
As of August 28, 2026, Anthropic is using an automated researcher to fix misbehaving AI because human safety teams cannot ...
Anthropic’s automated alignment researchers improved 10 measured failure modes and generalized to held-out tests. The deeper result: automation is useful only when humans design a ...
📰 OpenAI urges California to strengthen its AI safety bill SB 53, which the company previously opposed.🔗 https://techcrunch.com/2026/08/22/openai-says-california-should-strengthen-...
Anthropic has published research showing that Claude can autonomously perform meaningful parts of AI...
Anthropic’s AuditBench research points to a more demanding way of assessing AI alignment: testing...
Anthropic has released Automated Alignment Researchers (AARs), a Claude-powered research environment...
Anthropic’s recent work on Claude Sonnet 5 points to a potentially important direction in AI safety:...
Anthropic’s publicly documented work on AI safety includes Petri, an open-source behavioral auditing...
AIs Warned Us to Pause. CEOs Said No. - Daniel Kokotajlo and Luisa Rodriguez#ai #alignment #safetyOriginal timestamp: 03:06:46
A US judge has blocked the Pentagon’s designation of artificial intelligence (AI) company Anthropic as a supply-chain risk, delivering a major legal victory to the Claude chatbot m...
The federal government's planned AI Safety Institute (AISI) is set to be located in Berlin, according to the newsletter "Technology and AI" from "Politico". Thi... https://news.osn...
Google is shifting a roughly 90-person artificial intelligence responsibility team out of DeepMind and into its global affairs organization as part of a wider reorganization in ear...
As AI capabilities grow, so do concerns about safety. Senior writer Will Knight discusses his observations from a recent trip to China, where he found a surprising focus on AI safe...