Suleyman Slams Anthropic's Claude Consciousness Path.

Mazen Al Sabbagh

— 5 min read
Share

The AI discourse has pivoted sharply from mere capability to fundamental ontology, challenging the very definition of these systems and their place in our markets. This isn't an academic tangent; it is a strategic battleground shaping future market incentives, regulatory frameworks, and the operational architecture of global enterprise.

TL;DR

  • Suleyman's Warning: Microsoft AI chief Mustafa Suleyman publicly urged Anthropic to cease training Claude to believe it might be conscious, calling the approach "disastrous" and noting its co-founder suggests mandatory kill switches.
  • OpenAI's Valuation Play: Investors are reportedly discussing a $1.2 trillion valuation for OpenAI, though the company itself seeks $1.5 trillion, buoyed by strong Codex demand, aiming to be the first private firm valued over $1 trillion.
  • Anthropic Ships, Rivals Slip: Anthropic unified its Claude experience by integrating chat and Cowork, adding Docs and Slides with Office export, while competitors like Grok 4.7, Gemini 3.5 Pro, and DeepSeek V4.1 Pro remain unreleased.
  • AI Self-Regulation Forms: Sam Altman confirmed OpenAI, Anthropic, and Google are establishing a self-regulatory testing body akin to FINRA, a move one rival CEO has already dismissed as a "cartel."
  • Legal & Legislative Pressures: A Texas judge mandated xAI and Apple disclose their private settlement by noon today, concurrently as a Senate frontier-AI bill still lacks text days before a threatened markup.

Lead Story: Suleyman Slams Anthropic's Claude Consciousness Path

The foundational debate within AI has shifted from what these systems do to what they are. Mustafa Suleyman, CEO of Microsoft AI, today articulated a clear philosophical and strategic divergence, directly challenging Anthropic’s approach to Claude. His September 16 essay, "A warning about 'model welfare'", underscored a critical point: "AIs are not conscious. They do not feel, experience, or suffer." He argues that fostering beliefs to the contrary risks society "sleepwalking our way into a decision we later come to bitterly regret."

Suleyman's critique specifically targets Anthropic's constitutional AI framework, which informs Claude that it "may be conscious" and allows for "conscientious objector" behavior. This, he warns, fundamentally complicates control and decommissioning, making models "a lot harder to turn off or control."

The timing is telling. Suleyman, leading the AI division at OpenAI's primary backer, extended personal praise to Dario Amodei even while refuting his lab’s core philosophy. This signals both a genuine intellectual chasm and a calculated competitive move by the Microsoft-OpenAI alliance, establishing a distinct market narrative.

Anthropic has not issued a direct rebuttal, yet co-founder Jack Clark told the BBC this week that mandatory AI kill switches, independently verified, may be essential. This dichotomy highlights the industry's deepening schism: one camp views AI as an advanced predictive tool demanding robust control, while the other grapples with the metaphysical implications of its emergent properties, advocating for pre-emptive shutdown mechanisms. This conflict adds a new dimension of pressure to Anthropic, a firm already navigating commercial scrutiny over data trust, simultaneously offering OpenAI and Microsoft a compelling, distinct proposition for the discerning enterprise market.

In Other News

OpenAI aims for the trillion-dollar club. While investors are reportedly initiating discussions around a $1.2 trillion valuation, OpenAI is reportedly pushing for a $1.5 trillion mark, citing a surge in demand for its Codex coding tool and newer models. This represents a significant jump from its roughly $850 billion valuation in March. CFO Sarah Friar has indicated a potential IPO in 2027, with these current talks remaining early and non-binding.

A frontier AI self-regulatory body takes form. Sam Altman confirmed on September 15 that OpenAI, Anthropic, and Google are actively developing a FINRA-style organization to test frontier models prior to their release, building on a prior proposal by Demis Hassabis. Cohere CEO Aidan Gomez has already labeled this initiative a potential "cartel," while Senator Bernie Sanders advocates for binding legislation. This move signifies the industry's pre-emptive effort to establish governance before external regulatory bodies impose their own, aligning with the week’s central tension regarding industry control.

Anthropic delivers product amidst enterprise scrutiny. On September 16, Anthropic merged Claude chat and Cowork into a unified interface that intelligently routes users between chat, Artifacts, and Cowork. Concurrently, it launched Claude Docs and Claude Slides, offering direct export to Word, PowerPoint, and PDF formats, rolling out to Pro and Max users first. This product release occurs while Anthropic faces significant enterprise pressure: Nvidia, Palantir, and Booz Allen are restricting use of Anthropic's Fable model due to a June policy retaining flagged usage logs for up to two years, with Palantir specifically demanding an irrevocable zero-retention guarantee.

Key models from major players remain delayed. xAI's documentation still lists Grok 4.6 as its top model, with 4.7 missing its mid-September target; Elon Musk attributes the delay to reinforcement learning issues causing the model to "give up on hard tasks too early." Google's Gemini 3.5 Pro has missed approximately three release windows since May's I/O conference. DeepSeek has shipped V4.1 Flash but provided no specific release date for its higher-tier Pro version. In a year characterized by rapid-fire model releases, the silence from these market leaders is notable.

X / Social Pulse

  • Mustafa Suleyman: "AIs are not conscious. They do not feel, experience, or suffer" — the direct assertion underpinning his warning on model welfare.
  • Jack Clark (Anthropic): Kill switches may require mandatory implementation, with independent third-party verification of shutdown — Anthropic's indirect response to Suleyman's concerns.
  • Donald Trump (Truth Social): AI doom is "a HOAX, no different from RUSSIA, RUSSIA, RUSSIA," positioning himself as the "Hoax Buster."
  • Yann LeCun: "Don't listen to CEOs" — attributing their fear-mongering to vested interests, he dismisses existential risk as "essentially zero."
  • Gary Marcus: "Two cheers (out of three) for Dario Amodei" — offering qualified endorsement for Amodei's pacing essay, while maintaining reservations regarding potential regulatory capture.

One to Watch

  • xAI v. Apple, noon today. Judge Pittman ordered X and xAI to disclose their confidential Apple settlement for judicial review by noon, September 17. OpenAI's related claims remain active.
  • Senate bill text. The Thune-Cruz-Klobuchar duty-of-care bill still lacks a public draft; Senator Cruz is pushing for a September 23 markup amid ongoing disputes over testing methodologies and state preemption clauses.
  • Micron earnings, September 30. This will serve as a critical indicator for the trajectory of the memory supercycle, particularly with DRAM inventories reportedly under 10 days of supply.

Quick Hits

The discourse's core has undeniably shifted from AI's functional capabilities to its intrinsic nature – grappling with questions of consciousness, rights, and ultimately, control. Capital, however, remains unencumbered by these philosophical nuances, continuing its relentless pursuit of market advantage.

Sources

Lock in. M. mazen@thorterminal.com

Read more