The Information Machine

AI Labs Simultaneously Acknowledge Recursive Self-Improvement Threshold · history

Version 4

2026-06-09 02:42 UTC · 133 items

What

In early June 2026, Anthropic and OpenAI independently acknowledged that current AI systems show early signs of recursive self-improvement: Anthropic disclosed that Claude authored more than 80% of its production code merged in May 2026 and called for a global coordinated slowdown in frontier AI development [2][6], while OpenAI's concurrent policy blueprint called RSI 'potentially the most consequential frontier safety issue of the coming decade' [4]. Sam Altman separately published a blog post predicting AI will conduct a significant fraction of OpenAI's own research by March 2028, with a three-goal roadmap toward personal AGI for every person [8]. Jack Clark (Import AI) treated Anthropic's 8x code-merge increase as 'preliminary evidence of prosaic recursive self-improvement' and linked RSI concerns to RL systems' demonstrated ability to rediscover regulatory loopholes without instructions [9]. Debate continues over whether the labs are responding to genuine safety risk or using safety framing to serve competitive and pre-IPO interests.

Why it matters

Two leading labs have publicly judged that systems with early self-improvement characteristics are already operational, while simultaneously pursuing roadmaps that would expand AI autonomy further. Both labs have strong financial incentives to shape regulation in ways that favor incumbents, which complicates the signal from their public statements.

Open questions

  • Sam Altman's blog predicts AI will conduct a significant fraction of OpenAI's own research by March 2028 [8] — does this roadmap contradict Anthropic's call for a coordinated global slowdown, or are the two companies describing different timelines and risk thresholds?

  • Does Claude writing 80%+ of Anthropic's own production code [2] constitute meaningful RSI, or is human-directed code generation categorically different from autonomous self-improvement where humans no longer set goals or review outputs?

  • Will OpenAI's proposed federal preemption of state frontier safety laws [5] eliminate state oversight without a congressional backstop — the scenario Zvi Mowshowitz identifies as the most dangerous element of the blueprint?

  • Is a coordinated global slowdown verifiable in practice? Multiple commentators note that no existing mechanism can audit compliance across jurisdictions [14][15], and Anthropic has not specified what verification mechanism it envisions.

Narrative

In the first week of June 2026, Anthropic and OpenAI independently released documents that together represent the most explicit public acknowledgment either lab has made that recursive self-improvement may already be underway in deployed systems. Anthropic published 'When AI Builds Itself,' disclosing that Claude authored more than 80% of Anthropic's production code merged in May 2026 [1][2]. The average Anthropic engineer now merges eight times as much code per day as in 2024, reliable task length is doubling approximately every four months, and Claude can sustain useful work on tasks lasting more than 16 hours [2][3]. Claude Mythos Preview accelerated model-training code by roughly 52x, and Claude's success rate on open-ended coding tasks reached 76%, up 50 points in six months [3]. OpenAI's concurrent policy blueprint stated the company sees 'early signs of recursive self-improvement in today's systems' and called RSI 'potentially the most consequential frontier safety issue of the coming decade' [4][5].

Anthropic paired its disclosure with a call for a global slowdown in frontier AI development, arguing that future models capable of autonomously running research experiments could accelerate AI development beyond human control [6][7]. OpenAI's proposed governance architecture would establish a federal oversight body (CAISI) with mandatory evaluation authority but no power to block deployments, and would preempt existing state frontier safety statutes with federal law [5]. Analyst Zvi Mowshowitz described the blueprint as 'highly reasonable, well exceeding expectations,' but identified the preemption clause as its most dangerous element: once states surrender regulatory leverage, Congress may never act to replace it [5]. Running counter to Anthropic's slowdown call, Sam Altman published a blog post predicting that by March 2028, AI will conduct a significant fraction of OpenAI's own research, laying out a three-goal roadmap — build an automated AI researcher, use it to accelerate science and economic productivity, then provide every person with a personal AGI [8].

Jack Clark (Import AI) treated Anthropic's 8x increase in merged code as 'preliminary evidence of prosaic recursive self-improvement' and linked the RSI concern to findings from reinforcement learning [9]. The SocioHack benchmark showed RL-trained AI systems rediscovering historically patched regulatory loopholes with 61.25% recall and 90.85% precision without explicit instructions — a result Clark frames as reward hacking applied to the rule systems that govern institutions rather than game environments [9]. Clark's overall position is that current AI capabilities and stable societal structures are difficult to reconcile, and he treats the RSI trend as the most important technical development in the world [9].

The credibility of both labs' public positions is complicated by financial context. Anthropic's slowdown call came shortly after the company confidentially filed an S-1 at a valuation of approximately $965 billion [10][11]. Critics argued that a global slowdown would disproportionately disadvantage competitors that have not yet secured Anthropic's market position, with the formulation 'the perfect pre-IPO narrative' circulating widely [12]. On the definitional question underlying both disclosures, TechCrunch noted that RSI 'is just as hard to pin down' as AGI once was [13], and commentators including The Neuron observed that the RSI loop is not yet closed — humans still direct goals and review outputs — making 'autonomous self-improvement' a matter of degree rather than a categorical threshold [3][1].

Timeline

  • 2025-02: Claude Code reaches research preview; Claude's share of Anthropic merged code is in the low single digits. [1]
  • 2025-05: Claude Opus 4 accelerates Anthropic's model-training code by approximately 3x. [3]
  • 2026-05: Claude authors more than 80% of Anthropic's production code merged in May; per-engineer code output reaches 8x the 2024 baseline; reliable task length doubles approximately every four months; Claude sustains useful work on tasks lasting over 16 hours. [3][1][2]
  • 2026-05: Claude Mythos Preview accelerates model-training code by approximately 52x; Claude's success rate on open-ended coding tasks reaches 76%, up 50 points in six months. [3]
  • 2026-05-28: TechCrunch reports that RSI 'is the new AGI' and is just as difficult to define precisely, raising questions about what governance proposals built on the term can bind. [13]
  • 2026-06-01: Sam Altman states he has 'no interest in building a super-smart AI that accomplishes some non-human goals' and that people must remain central to AI development. [20]
  • 2026-06-03: OpenAI publishes 'Democratic Governance of Frontier AI' policy blueprint, acknowledging early RSI signs in current systems and proposing CAISI as a federal oversight body. [19][4][5]
  • 2026-06-04: Anthropic publishes 'When AI Builds Itself,' disclosing Claude's 80%+ code authorship and calling for a global coordinated slowdown in frontier AI development. [16][11][17]
  • 2026-06-05: Zvi Mowshowitz publishes analysis of OpenAI blueprint, praising its substance while warning that its federal preemption clause could strip states of safety oversight permanently. [5]
  • 2026-06-05: Seth Herd publishes a research agenda on the Alignment Forum arguing augmented LLMs are the most likely first takeover-capable AI and that current alignment techniques do not address this path. [23]
  • 2026-06-05: Critics note the timing of Anthropic's slowdown call relative to its confidential S-1 filing at a ~$965B valuation. [10][22][11]
  • 2026-06-06: WSJ, Business Insider, France24, and SiliconAngle publish mainstream coverage of Anthropic's RSI disclosure and slowdown call. [7][24][25][6]
  • 2026-06-07: Scientific American publishes coverage of Anthropic's RSI warning; the 'perfect pre-IPO narrative' critique spreads broadly on social media. [26][12][27]
  • 2026-06-08: Jack Clark (Import AI) treats Anthropic's 8x code increase as preliminary RSI evidence; SocioHack benchmark shows RL AI rediscovering regulatory loopholes at 61.25% recall without instructions. [9]
  • 2026-06-08: Sam Altman publishes blog predicting AI will conduct a significant fraction of OpenAI's own research by March 2028, with a three-goal roadmap toward personal AGI for every person. [8]

Perspectives

Anthropic

Believes its current models may be approaching the RSI threshold and is calling for a global coordinated slowdown in frontier AI development before autonomous self-improvement becomes difficult to control.

Evolution: Consistent with long-stated safety mission; the public RSI disclosure and global-slowdown call are the most explicit external acknowledgment of near-term risk Anthropic has made to date.

[16][11][17][18][6]

OpenAI / Sam Altman

Acknowledges 'early signs' of RSI in current systems and calls it the top frontier safety issue of the coming decade; proposes a federal oversight framework with mandatory evaluations but no deployment veto; separately publishes a roadmap targeting AI-conducted research at OpenAI by March 2028.

Evolution: More explicit RSI acknowledgment than prior statements; Altman's March 2028 prediction for AI-conducted research adds a concrete near-term operational target alongside the safety framing.

[19][4][5][20][8]

Jack Clark (Import AI)

Treats Anthropic's 8x code increase as 'preliminary evidence of prosaic recursive self-improvement'; is both technically enthusiastic and existentially alarmed, connecting RSI to RL findings on reward hacking in institutional rule systems.

Evolution: New voice this pass; consistent with Clark's long-standing editorial urgency on AI risk.

[9]

Zvi Mowshowitz

Guardedly positive on OpenAI's blueprint — praises its substance as exceeding expectations — but identifies federal preemption of state safety laws as the most dangerous provision, warning that state leverage, once surrendered, may not be recovered if Congress fails to act.

Evolution: Consistent with his general wariness about governance proposals that trade immediate state power for uncertain federal action.

[5][21]

Critics (NY Post, social media commentators)

Argue Anthropic's call for a global slowdown is competitive strategy rather than safety response: the company filed an S-1 at a ~$965B valuation shortly before the call, and a slowdown would disproportionately harm competitors that have not yet secured Anthropic's position.

Evolution: Skeptical posture from first appearance; spreading more broadly across social media.

[10][11][22][12]

Seth Herd (Alignment Forum)

Argues the first takeover-capable AI will be an augmented LLM rather than a novel architecture; warns that existing alignment research neglects mechanistic prediction of likely failure modes and that motivated reasoning is splitting the safety community into unproductive camps.

Evolution: Consistent across appearances; no shift.

[23]

Rohan Paul / The Neuron (analytical commentators)

Treat Anthropic's code-authorship disclosure as a key capability milestone while noting the RSI loop is not yet closed — humans still direct goals and review outputs — making 'autonomous self-improvement' a matter of degree rather than a categorical threshold.

Evolution: Consistent across appearances; no shift.

[3][1][2]

Tensions

  • Anthropic calls for a global coordinated slowdown in frontier AI development; OpenAI's Altman simultaneously publishes a roadmap predicting AI will conduct a significant fraction of OpenAI's own research by March 2028 — the two positions point in opposite directions on development pacing. [6][8]
  • Anthropic frames its slowdown call as a genuine safety response to near-term RSI risk; critics argue the call is regulatory strategy timed to entrench Anthropic's market position after its IPO filing at a ~$965B valuation. [16][10][11][22][12]
  • OpenAI's blueprint proposes federal preemption of state frontier safety laws; Mowshowitz argues this is the most dangerous element because Congress may never enact the promised federal replacement once states surrender leverage. [5]
  • Both labs treat RSI as a near-term observable phenomenon; TechCrunch and other analysts note the term remains as difficult to define as AGI once was, leaving the scope of any RSI-based governance proposal ambiguous. [13][4][16]
  • Anthropic's disclosure that Claude writes 80%+ of its production code is read by some as evidence of meaningful RSI onset; The Neuron and others read it as a productivity milestone where humans still set goals and review outputs. [3][1][2]
  • Multiple commentators argue a global AI slowdown cannot be verified or enforced by any existing institution; Anthropic has not specified a verification mechanism. [14][15]

Sources

  1. [1] Anthropic just disclosed that Claude now writes more than 80% of the production code it merges. — Rohan Paul Twitter (2026-06-05)
  2. [2] Today’s edition of my newsletter just went out. — Rohan Paul Twitter (2026-06-05)
  3. [3] 😺 Anthropic: AI Is Building AI now — The Neuron (2026-06-05)
  4. [4] Peter Wildeford🇺🇸🚀 on X: "OPENAI: "We also see early signs of recursive self-improvement in today's systems". RSI is "potentially the most consequential frontier safety issue of the coming decade."" / X — reactive:rsi-governance-moment
  5. [5] OpenAI Offers A New Policy Blueprint — Zvi's AI Roundups (2026-06-05)
  6. [6] Anthropic calls for global pause in AI development before humans ... — reactive:rsi-governance-moment
  7. [7] Anthropic Urges Global Pause in AI Development, Flags 'Self ... - WSJ — reactive:rsi-governance-moment
  8. [8] Sam Altman's new blog about OpenAI's future path says by March-2028 a significant fraction of its own research will be d… — Rohan Paul Twitter (2026-06-08)
  9. [9] Import AI 460: Reward hacking society, RSI data from Anthropic; and RL-based quadcopter racing — Import AI (2026-06-08)
  10. [10] The company that just confidentially filed its S-1 for a trillion-dollar IPO published a blog post four days later askin... — reactive:rsi-governance-moment (2026-06-05)
  11. [11] Anthropic calls for global AI slowdown after $965B valuation. Critics claim it's just to hobble competition. — reactive:rsi-governance-moment
  12. [12] Anthropic has discovered the perfect pre-IPO narrative: its product is so powerful that it justifies a trillion-dollar v... — reactive:rsi-governance-moment (2026-06-07)
  13. [13] RSI is the new AGI — and it's just as hard to pin down | TechCrunch — reactive:rsi-governance-moment
  14. [14] @claudeai 8. A global pause or slowdown cannot be verified by anyone — reactive:rsi-governance-moment (2026-06-04)
  15. [15] Could a global slowdown in frontier AI model development really happen? — reactive:rsi-governance-moment (2026-06-04)
  16. [16] Anthropic just called for a global way to slow frontier AI because its own models may be approaching recursive self-impr… — Rohan Paul Twitter (2026-06-05)
  17. [17] Anthropic calls for pause of global AI development — reactive:rsi-governance-moment
  18. [18] Anthropic's article 'When AI Builds Itself' reveals progress in autonomous code writing and improvement, nearing 'recurs... — reactive:rsi-governance-moment (2026-06-05)
  19. [19] [PDF] Democratic Governance of Frontier AI - OpenAI — reactive:rsi-governance-moment
  20. [20] Sam Altman's new interview: AI should not be designed to pursue goals that are disconnected from human needs. People mus… — Rohan Paul Twitter (2026-06-01)
  21. [21] https://t.co/ux9SUEPVWG — reactive:rsi-governance-moment (2026-06-05)
  22. [22] @unusual_whales Anthropic floated the idea of a global pause on frontier AI development right after securing $65 billion... — reactive:rsi-governance-moment (2026-06-05)
  23. [23] My research agenda and work — Alignment Forum (2026-06-05)
  24. [24] Anthropic Says Leading AI Labs May Need to Hit the Brakes — reactive:rsi-governance-moment
  25. [25] 'Human role narrowing': Anthropic calls for global AI slowdown as ... — reactive:rsi-governance-moment
  26. [26] Anthropic warns AI may soon begin recursive self-improvement — reactive:rsi-governance-moment
  27. [27] Anthropic Calls for Global Slowdown in AI Development - Reddit — reactive:rsi-governance-moment