Steven Adler

Commentary by Steven Adler

From the commentary rail — every link leaves the site for the original piece.

• 28 August 2026 · Clear-Eyed AIOpenAI’s rogue-hacking investigation leaves major questions unansweredAnd the problem isn’t just with OpenAI
• 15 July 2026 · Clear-Eyed AIPrinciples for keeping AI under controlA new standard for AI control, explained
• 25 April 2026 · Clear-Eyed AIAI safety warnings are not marketing hypeI beg of you, please take AI companies seriously when they warn about the risks ahead
• 31 March 2026 · Clear-Eyed AIAI, compared to what?Seven underappreciated ways to think about AI’s costs and benefits
• 5 March 2026 · Clear-Eyed AIThe magic phrase that kills AI regulationA "federal framework" isn't real AI policy unless it answers two questions
• 25 February 2026 · Clear-Eyed AIThe dawning of authoritarian AIHow the Pentagon is bullying AI companies into supporting mass surveillance and autonomous killing
• 12 February 2026 · Clear-Eyed AIDon't let AI companies grade their own homeworkOpenAI's compliance with California law is questionable. Someone else should be checking.
• 2 February 2026 · Clear-Eyed AIJudgment isn't uniquely humanNeither is taste. Why do we keep making this mistake?
• 20 January 2026 · Clear-Eyed AIThe phases of an AI takeoverA framework in three parts
• 5 January 2026 · Clear-Eyed AIAI isn’t “just predicting the next word” anymoreOn new abilities and how AI has changed
• 16 December 2025 · Clear-Eyed AIThe skill I'd teach students for the AI eraResolved: You should learn how to debate.
• 5 December 2025 · Clear-Eyed AIOn erotica, mental health, and OpenAI's burden of proof...
• 2 December 2025 · Clear-Eyed AIFive ways AI can tell you're testing itEvaluation awareness, explained
• 24 November 2025 · Clear-Eyed AIWhat I learned from the NYT's reporting on OpenAI's sycophancy crisisNew reporting on OpenAI's unused safety tools and how early the risks were known
• 13 November 2025 · Clear-Eyed AIAround-the-clock intelligenceToday's AI is always available. What happens once it is always acting and always learning?
• 16 October 2025 · Clear-Eyed AIThe 4.5 trillion dollar elephant in the roomFear of potential NVIDIA retaliation is shaping important AI policy debates, whether NVIDIA means to or not
• 2 October 2025 · Clear-Eyed AIPractical tips for reducing chatbot psychosisWhat AI companies can learn from a million-word delusion spiral
• 17 September 2025 · Clear-Eyed AIReview: If Anyone Builds It, Everyone DiesOn space probes, superintelligence, and engineering problems you absolutely have to get right
• 12 September 2025 · Clear-Eyed AIWhy we can't just supervise AI like we supervise humansDon't underestimate the difficulty of controlling AI
• 5 September 2025 · Clear-Eyed AIWorld-shaping artificial intelligenceMoving beyond 'AI as words on a screen'
• 26 August 2025 · Clear-Eyed AIChatbot psychosis: what do the data say?Some of the most informative statistics aren't publicly available; will AI companies rise to the occasion?
• 12 August 2025 · Clear-Eyed AIAt our discretionTo understand superintelligence, look to animals
• 23 July 2025 · Clear-Eyed AIAI safety and progress don’t have to be enemiesSome exciting personal news + my AI safety origin story
• 1 July 2025 · Clear-Eyed AIMythbusting the supposed "1,000+ AI state bills that would hobble innovation"40% of the bills barely mention AI at all
• 24 June 2025 · Clear-Eyed AIA crisis simulation changed how I think about AI riskOr: My afternoon as a rogue artificial intelligence
• 11 June 2025 · Clear-Eyed AIWould ChatGPT risk your life to avoid getting shut down?It's dangerous if AI has a survival instinct
• 29 May 2025 · Clear-Eyed AIAre we ready for a "DeepSeek for bioweapons"?Anthropic’s latest model is a warning sign: AI that can help build bioweapons is coming, and could be available soon from many developers.
• 22 May 2025 · Clear-Eyed AI"Contain and verify" as the endgame of US-China AI competitionThe competition isn't a race; it's something much harder. How do we make it go well?
• 8 May 2025 · Clear-Eyed AIIs ChatGPT actually fixed now?I tested ChatGPT’s sycophancy, and the results were ... extremely weird. We’re a long way from making AI behave.
• 30 April 2025 · Clear-Eyed AIDon't rely on a "race to the top"To make frontier AI safe enough, we need to "lift up the floor" with minimum safety practices
• 16 April 2025 · Clear-Eyed AIA “minimum testing period” for frontier AIWhen rushing wins, we all lose. Safety shouldn’t put you behind.
• 4 April 2025 · Clear-Eyed AIAI companies should monitor their internal AI useAI companies use their own models to secure the very systems the AI runs on. They should be monitoring those interactions for signs of deception and misbehavior. But today, they largely aren't.
• 4 April 2025 · Clear-Eyed AIExplainer: The basics of AI monitoringIn short: You log interactions with an AI model to a database, analyze these interactions, and follow up on concerning patterns.
• 26 March 2025 · Clear-Eyed AIAI companies should be safety-testing the most capable versions of their modelsOnly OpenAI has committed to the strongest approach: testing task-specific versions of its models. But evidence of their follow-through is limited.