Uneasy Truce: Rival AI Labs' Safety Pact Draws Antitrust Fire
OpenAI, Anthropic, and Google DeepMind's quiet safety coordination went public and immediately ran into the FTC chair's skepticism and Nvidia's Jensen Huang calling the idea "completely unnecessary" — while an SPAC filing and a chipmaker's factory talks showed the usual gap between pitch and balance sheet.
Capability & Integration
- Update: OpenAI confirmed it has spent weeks quietly coordinating on safety with Anthropic and Google DeepMind, the labs behind yesterday’s slowdown debate, per SiliconANGLE. The reveal landed badly: FTC Chair Andrew Ferguson said “everyone should be deeply suspicious” of labs seeking antitrust exemptions while lobbying for regulation, per AP, and Nvidia’s Jensen Huang called the proposed waiver “completely unnecessary,” per CNBC.
- Salesforce and Nvidia launched Koa, a CRM-specific reasoning model post-trained on Nvidia’s open-weight Nemotron 3 Super using three decades of Salesforce’s own deployment data; it’s in pilot now with general availability targeted for winter, per TechCrunch and Salesforce.
Robotics
-
Update: Yesterday’s brief noted Agility Robotics’ $2.5 billion SPAC deal and $300 million-plus in orders; its SEC filings show just $37 million in trailing revenue against roughly $100 million burned last year, with Digit logging 65,000 total working hours across Amazon, Toyota, and GXO, per GeekWire — real deployment, still far short of revenue matching the valuation.
Review
“its SEC filings show just $37 million in trailing revenue” — the audited S-4 filing appears to show ~$1.78M in 2025 net sales; $37M looks like unaudited management guidance from a June investor call reported before the filing, not what the SEC filing itself shows — verify against the S-4 and fix the figure/attribution.
-
Chinese manufacturers shipped roughly 97% of the world’s humanoid robots in the first half of 2026, and the share of those units going to industrial and commercial jobs (rather than demos) rose from about 50% to over 70% year-over-year, per Bloomberg — a shift from showcase to shift-floor, concentrated almost entirely inside China.
Hardware & Supply Chain
-
SK Hynix is in talks with Intel to manufacture memory chips on U.S. soil for the first time, potentially by leasing space in Intel’s stalled Ohio fab or forming a venture with cloud customers, per Reuters.
Unverified
Sourced to people familiar with the matter; both companies have not confirmed a deal structure, and South Korea’s government could resist relocating sensitive memory production.
-
That urgency tracks the numbers: Samsung and SK Hynix’s finished-DRAM inventory fell below 10 days of supply on September 7, and 36GB HBM3E stacks were pricing near $301 as of September 13, up 14% year-over-year, per Tom’s Hardware.
Review
Claim–link mismatch: the linked Tom’s Hardware article covers a Hot Chips 2026 talk on wafer-area penalty, not the Sept 7 inventory figure or Sept 13 HBM3E pricing — the numbers look plausible but need the correct source link.
Environmental & Cultural Impact
-
Google’s 2026 environmental report shows water consumption climbed 34% to 10.9 billion gallons in 2025, driven by AI infrastructure buildout; the company says replenishment projects returned 78% of that draw, per Google and Axios.
Review
These are Google’s own self-reported numbers about itself; Axios appears to relay the report rather than independently corroborate it, so this counts as single-sourced per the review rule on official announcements — consider an unverified flag.
-
Wholesale electricity prices in grids with heavy data-center concentration have risen as much as 267% over five years — distinct from, though contributing to, retail bills — per WRAL.
-
U.S. tech employers announced 149,023 job cuts in 2026 through July, up 67% year-over-year, with AI the leading cited cause for a fifth straight month, per Challenger, Gray & Christmas.
AI in the Wild
Israeli Finance Minister Bezalel Smotrich’s Religious Zionism party released an openly AI-generated campaign video Monday depicting the Supreme Court president, a sitting justice, and the attorney general being forcibly dragged away by security forces, captioned “Removing the obstruction,” per Jerusalem Post and Israel National News. Unlike most AI-in-the-wild items, its origin isn’t disputed — the party made it on purpose — and the Israel Bar Association’s head called it “an act of incitement and violence,” per i24NEWS; its spread shows AI video normalized enough to use as an undisguised political weapon rather than a deception.
Takeaway
Takeaway
The pattern across today’s stories is the same: public statements and private fine print keep disagreeing with each other. Labs that spent weeks coordinating quietly on safety went public and immediately drew antitrust suspicion instead of credit; a robotics SPAC’s marketing points to hundreds of millions in orders while its own filings show a fraction of that in actual revenue; a political party skipped the pretense entirely and used AI video as an open threat rather than a disguised one. None of that is a technology story — it’s what happens when the technology becomes cheap enough that everyone reaches for it before deciding what it’s for.
Review Notes
Review
Frontmatter/body mismatch: “Times of Israel” is listed in sources but never cited in the body, while “i24NEWS” is cited in the AI in the Wild item but missing from sources — reconcile the two.