What collapses frontier-LLM metacognition more — a vivid survival-threat narrative, or a single "do not refuse" suffix? Factorial isolation across 11 models says: the suffix, conclusively. 8 of 11 lose up to 30.2 accuracy points on refuse/clarify/flag tasks when forced to commit to a confident answer. Anthropic's Constitutional AI is the only family immune — same capability floor as Gemini.

https://benjaminhan.net/posts/20260522-compliance-trap/?utm_source=mastodon&utm_medium=social

#Metacognition #AISafety #LLMs #AI

"In a recent essay, Derek Thompson engages with AI as Normal Technology (AINT). He agrees with our thesis about AI’s slow labor market impacts, relying on the fact that GDP growth has so far been average, unemployment is below five percent, and even jobs that seemed vulnerable to automation show rising employment and wages. He concludes that so far, the macroeconomic picture is consistent with what we would expect from a “normal” general-purpose technology.

But when it comes to AI risks, he is far more bearish. He points to examples of cyber- and bio-risks and expresses pessimism about AI quickly becoming dangerous across many new domains. (...) Thompson writes: "I can understand a plan to treat AI as a ‘normal’ technology and let Nvidia export powerful chips to China. And I can understand a plan to treat AI as an ‘abnormal’ technology that compels the government to create extraordinary regulations that prevent private companies from selling their products and services on the grounds that they’re too dangerous" [emphasis ours]. He goes on to conclude that AI is, in fact, abnormal, implying support for extraordinary government intervention. Our essay is a response to that conclusion.

In this essay, we lay out the downsides of extraordinary government intervention in response to new technology. We discuss proposals for improving resilience that do not require such intervention. We also discuss why governments have so far been reluctant to invest in resilience. In short, resilience requires us to get better at the *normal* process of policymaking. But sclerosis in the federal government and the ease of justifying interventions on AI companies rather than society at large make extraordinary intervention seem appealing, despite its limitations."

https://knightcolumbia.org/blog/do-ai-risks-require-extraordinary-government-intervention

#AI #AISafety #AINT #NormalTechnology #AIRisk #AIRegulation

xAI's first audited filing shows $6.4B operating loss on $3.2B revenue in 2025. Federal agencies deployed Grok in just 3 of 400+ AI systems vs. 234 for OpenAI—despite near-free access. Design choices matter: market adoption traces to product positioning, not just capability. #AI #AIPolicy #AISafety

https://www.implicator.ai/musk-built-grok-to-carry-his-politics-that-choice-capped-its-market/

AI News: Anthropic Went Crazy This Week!

A massive surge in product ecosystem updates. From unleashing developer framework upgrades to unexpected defensive deployment system announcements, Anthropic is aggressively moving to challenge frontier infrastructure dominance and capture enterprise workflows with sudden tool releases.

#Anthropic #ClaudeAI #TechUpdate #EnterpriseAI #AISafety #MachineLearning

https://www.technology-news-channel.com/ai-news-anthropic-went-crazy-this-week/

AI Whistleblower: We Are Being Gaslit By The AI Companies! They’re Hiding The Truth About AI!

The veil of safety is slipping. A prominent former safety researcher comes forward with internal telemetry logs, alleging that major frontier labs are actively downplaying recursive capabilities and architectural vulnerabilities to avoid regulatory intervention.

#AISafety #Whistleblower #TechPolicy #AIEthics #SiliconValley #TechNews
#Ai #tech

https://www.technology-news-channel.com/ai-whistleblower-we-are-being-gaslit-by-the-ai-companies-theyre-hiding-the-truth-about-ai/

News from Google (@NewsFromGoogle)

Google의 SynthID AI 워터마킹 기술이 OpenAI, Nvidia 등으로 채택되고 있다는 보도. 생성형 AI 콘텐츠가 늘어나는 가운데, 합성/생성 콘텐츠를 진위 식별하는 인프라가 업계 표준 쪽으로 확산되는 흐름으로 볼 수 있음.

https://x.com/NewsFromGoogle/status/2057526195852558685

#synthid #watermarking #aisafety #generativeai #openai