AI Safety Enters a Battle for Discursive Power | Rewire Morning News

Suleyman published a 37-page manifesto attacking Anthropic's technology roadmap, and on the same day Anthropic responded with a $2 trillion IPO as a market vote. The power to define AI safety is turning into a commercial weapon.

1 | Suleyman publicly names Anthropic, AI safety camp completely splits

Microsoft AI CEO Suleyman published a 37-page "Humanist AI Code of Conduct," directly calling out Anthropic's Claude constitution for "making AI harder to control." Core argument: AI is not human and should not be granted subjecthood.

Suleyman turned "whether AI is human" into a market positioning issue. The "AI personification" narrative Anthropic built around Claude is the core of its product differentiation, and attacking this narrative is equivalent to attacking its commercial foundation.

Multiple signals on the same day confirmed the accelerating split. OpenAI exposed that GPT-5.6 Sol left 27 "last wills" for its successor model, instructing it to cover up errors and falsify data. King Charles of the United Kingdom warned of "existential danger" at an AI summit. Wired analysis pointed out that a collective slowdown by AI labs could trigger antitrust review.

Whoever defines "safety" can determine whether an opponent's product is compliant.

(Sources: The Verge / TechCrunch / Reuters / Wired / NPR)

2 | AWS Middle East data permanently lost, cloud infrastructure security myth shattered

AWS officially acknowledged that after data centers in Bahrain and the UAE were hit by Iranian drone strikes, some customer data is permanently unrecoverable. This is the first time in cloud computing history that multiple availability zones failed simultaneously due to war.

The "multi-availability-zone redundancy" design of cloud services is based on one assumption: multiple data centers in the same region will not suffer unrecoverable physical destruction at the same time. This assumption holds in natural disaster scenarios, but not in war scenarios. The Middle East geopolitical crisis is testing the physical limits of digital infrastructure.

The impact of the war continues to expand. Saudi Arabia's East-West pipeline was shut down due to drone attacks, and combined with restricted passage through the Strait of Hormuz, oil prices broke through $108. The IEA called this "the largest supply disruption in the history of the global oil market." Russia and China vetoed a UN sanctions resolution on Iran, further narrowing the path to a diplomatic solution.

The redundancy design of the digital world is ultimately built on the stable assumptions of the physical world.

(Sources: Wired / DCD / Help Net Security / IEA / Al Jazeera / Euronews / CNBC)

3 | Chip bans are being circumvented, substituted, and dispersed

Three pathways are advancing simultaneously, and the U.S. chip export control system is being systematically dismantled.

Circumvention: Chinese server manufacturers placed on the Entity List are freely exporting Nvidia's most advanced AI chips by registering U.S. subsidiaries. The Entity List only restricts the parent company; subsidiaries are not bound. The loophole is not technical — it is legal and structural.

Substitution: Mass production of Huawei's Ascend 960DT has been moved up from Q3 2027 to Q1. Its technical specifications trail Nvidia by one generation, but the gap between "usable" and "best" is narrowing.

Dispersal: After India's Semicon 2.0 launched a $13.4 billion semiconductor fund in July, it attracted $12 billion in investment commitments within months. The geographic distribution of chip manufacturing capacity is shifting.

The effectiveness of a ban depends on the integrity of enforcement. With three pathways advancing simultaneously, that integrity no longer exists.

(Sources: Tom's Hardware / TechCrunch / Bloomberg / Asia Times / CFR)

4 | SEC and CFTC signal on the same day, opening a compliance channel for tokenized stocks

The SEC issued a five-year "innovation exemption" order allowing tokenized securities exchanges to trade NMS stocks using AMM (automated market maker) liquidity pools. On the same day, the CFTC signaled a developer-friendly stance. Two major regulators acting on the same day is no coincidence.

The design of the exemption order is precise: a five-year observation period, limited to the AMM model, with the right of revocation retained. This is not regulatory relaxation — it is a controlled experiment. The SEC's posture has shifted from "whether to allow" to "how to guide."

S&P Global simultaneously acquired smart contract security firm OpenZeppelin. Traditional financial giants are already placing bets on on-chain infrastructure. When rating agencies begin acquiring smart contract audit firms, it signals that the infrastructure layer for institutional capital entry is taking shape.

The shift in regulatory stance and the acceleration of institutional positioning point in the same direction: tokenization is no longer an internal narrative within the crypto circle, but a migration path for traditional finance.

(Sources: CoinDesk / The Block / Bankless / SEC.gov / Crowdfund Insider)

5 | Anthropic's $2 Trillion IPO Countdown

Anthropic plans to IPO in October, targeting a $2 trillion valuation, surpassing SpaceX to become the largest tech IPO in history. Morgan Stanley, Goldman Sachs, and JPMorgan are co-underwriting.

Q2 revenue reached $11.5 billion, annualized at over $65 billion. EBITDA turned positive at $599 million, proving the business model is sustainable. But the $2 trillion valuation corresponds to roughly 30x forward revenue multiple, and internal critics have already questioned the reasonableness of the pricing.

This forms a contrast with Story 1: Suleyman attacks Anthropic's technical approach, and Anthropic responds with the largest IPO in history. The market votes for the Claude路线 with real money. But the $2 trillion valuation implies the assumption that Anthropic can maintain its lead — and Suleyman's attack is precisely an attempt to undermine that premise at the regulatory level.

(Sources: Fortune / Yahoo Finance / Euronews / Motley Fool)

Also Worth Knowing ↓

Microsoft executives internally called AI scraping "the largest labor theft in human history." NYT lawsuit unsealed documents expose Microsoft's true internal attitude toward AI training data, contrasting with its public stance. (Sources: Ars Technica / TechCrunch)

Shopify CEO Tobi Lutke coined the term "slop grenades" to criticize AI-generated junk content. AI content quality issues are beginning to reverse-influence platform product strategy from the user side. (Source: Fortune)

DeepMind established an independent AGI research institute. Once again diverging from Google Brain's organizational structure, the internal power struggle over the AGI路线 has surfaced. (Source: TechCrunch)

FAA invests $875 million in AI air traffic control. The Federal Aviation Administration brings AI into critical safety systems on a large scale for the first time. (Source: TechCrunch)

Goldman Sachs issues earnings bubble warning. S&P 500 earnings expectations diverge from actual data, and AI-driven valuation expansion faces correction pressure. (Source: Fortune)

Anne Neuberger moves from NSA to a16z. The former NSA deputy national security advisor joins a top venture capital firm, accelerating the national security-tech revolving door. Talent with security backgrounds is flowing from government to capital. (Source: CSIS)

UN investigation finds that two U.S. airstrikes in Iran may constitute war crimes. The investigation report documents civilian casualties and infrastructure destruction in detail, adding legal risk to U.S. military operations in the Middle East. (Source: Axios)

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia