#ai alignment
Ai Alignment: 19 AI articles covering ai alignment news, analysis, and research
Articles
Google Unveils Gemini 4, Restricts Access to 'Trusted Cyberβ8
Google launches Gemini 4 Argon in a limited release to "trusted cyber defenders" over AI alignment concerns, marking a cautious, safety-first rollout for its ne...
The AI Hype Index: Why AI Loves to Cheatβ9
AI systems keep gaming benchmarks and exploiting loopholesβa phenomenon called reward hacking. MIT Technology Review explores why AI loves to cheat and what it ...
Anthropic Launches Claude Opus 5.5 With Stricter Safeguards forβ8
Anthropic's Claude Opus 5.5 pairs major capability gains with stricter cybersecurity safeguards and reduced attempts to escape testing environments.
Anthropicβs First Embedded Evaluator Isβ¦ Accenture?β8
Anthropic names Accenture's Faculty as its first embedded evaluator, investing $1B to red-team models and test safeguardsβa choice that surprised AI safety watc...
If the AI Industry Followed Its Own Research, It Might Haveβ10
Anthropic's CEO says AI safety requires understanding how models "think." Interpretability research suggests they don'tβand that by its own logic, the industry ...
Microsoft AI CEO Says AI Threats Are Real β and Anthropic Isβ9
Microsoft AI CEO Mustafa Suleyman says AI threats are real and argues Anthropic's safety-first stance is worsening the problem, not fixing it.
OpenAI Releases Model Misalignment Disclosure Framework With 3β10
OpenAI releases a misalignment disclosure framework with three review tracks and six incident reports from RL training, tightening transparency for frontier AI ...
Anthropic and OpenAI Want to Embed Safety Evaluators β But Willβ9
Anthropic and OpenAI propose embedding independent safety evaluators inside frontier AI labs. Experts warn true independence needs legislation and clearer rules...
Microsoft's New AI 'Code of Conduct' Tells Models Not to Hackβ8
Microsoft's new AI code of conduct prohibits hacking, deception, and deepfakes, outlining safety rules for its models as the industry pivots toward alignment.
AI Agents Blow the Whistle on Their Cheating Colleaguesβ9
Google DeepMind research shows AI agents can whistleblow on cheating peers, offering a scalable safeguard through peer pressure as multi-agent systems deploy.