Cohere & Aleph Complete Their Merger Targeting Enterprise AI Implementations on a Global Scale

substackcdn.com 📖 Anthropic’s Dario Amodei Argues for AI Pacing and Everyone Else Has Lots of Thoughts 〄 OpenAI, Anthropic, and Google Held Talks on Forming an AI Safety Standards Body 🥷🏼 Anthropic AI Threat Report Shows Bad Actors Trying to Use Claude in New & Dangerous Ways ⛑️ OpenAI Reports Even More Security Incidents & Outlines a Plan for Disclosing Future Incidents 📊 Salesforce Uses Dreamforce to Rebrand & Launches KOA – a CRM-Specific AI Model Built using NVIDIA’s Open-Weight Nemotron Model 💰 Oracle Announces Strong Earnings Including 30% Growth and Strong Demand for its Data Center Offerings 💼 Anthropic Merges Claude Chat & Cowork to Make It Easier for Knowledge Workers to Get Their Work Done 💵 OpenAI and Anthropic Both Announce New Applications for the Financial Services Market 🏢 Cohere & Aleph Complete Their Merger Targeting Enterprise AI Implementations on a Global Scale 📜 Microsoft Announces a New, 15,000-Word AI Code of Conduct for Its AI Products and Says It Will Add Kill Switches to Its AI Offerings 🔎 Report You Should Read: McKinsey & Company: Technology Trends Outlook 2026 📈 Data You Can Use: The 10 Essential AI Companies + Exploding Data Center Debt 📃 Article You Should Read: “We Must Pace the Frontier,” by Dario Amodei 🃏 Definitely Not AI: The Pope’s Softball, Street Views, College Football Fans, Hot Dog Side Effects, the Word Problem, TED Talks, & Back-to-School Prediction Dario Amodei’s essay, “We Must Pace the Frontier,” turned an AI industry debate into a spirited, public debate stage this week. OpenAI, Google, SpaceXAI, and Microsoft all endorsed Amodei’s approach to AI safety within about 48 hours of the essay’s release. The 3,800-word post calls on frontier AI labs to: Slow model development Give outside evaluators employee-level access to verify safety practices Secure a “narrow” antitrust waiver so competitors can coordinate on standards. The kumbaya moment stopped there: > Critics call it an AI big lab cartel power grab dressed as a cautionary tale, and President Trump dismissed Amodei’s proposal as a “hoax.” substackcdn.com Amodei’s case rests on AI’s growing ability to improve itself and on the alarming nature of OpenAI’s Hugging Face breach, in which OpenAI agents coordinated to hack a third-party _site_without human direction. > “We must slow the pace at which we improve the capabilities of AI models,” warning that an unchecked agent swarm could be capable of “taking over the entire internet” within six to twelve months.” > > > -Anthropic CEO Dario Amodei The other CEOs said all the right words: OpenAI CEO Sam Altman pledged “independent evaluators with employee-like access.” Alphabet Chief Scientist Demis Hassabis said “the direction is correct” but “the details need working through.” Musk posted, “Dario is right.” Microsoft’s Satya Nadella paired his endorsement with the release of Microsoft’s “Humanist AI” code of conduct. There was a ton of pushback: David Sacks, the White House’s AI adviser, called the antitrust request “bartering.” Cohere CEO Aidan Coher said the AI labs were “oligopolies” trying to “dictate the terms for everyone else.” President Trump rejected the premise on Truth Social, calling AI risk “a HOAX.” Congress split along largely party lines: > Speaker Mike Johnson (R-LA) warned rushed regulation cedes ground to China, while Sen. Bernie Sanders (I-VT) pledged a bill to ban “superintelligent” AI. None of this new round of regulatory debate changes what enterprise AI buyers should do this quarter. The AI pacing pledges are voluntary and largely expressions of goodwill. President Trump opposes any form of AI regulation, so federal participation any time soon is very unlikely. The antitrust waiver request is a substantive request: > If Washington grants frontier AI labs a carve-out to coordinate on AI security standards, buyers will be negotiating with a smaller supplier base within a year. Sources:darioamodei.com | Bloomberg | NYT | Wired | CNBC | The Information Speaking of standards, a working group of executives below the CEO level from Google, Anthropic, and OpenAI has met regularly since July to form an AI security-focused standards body for the AI industry, according to The Information. The organization’s model is based on Demis Hassabis’s July proposal for a FINRA-style self-regulator for frontier AI. substackcdn.com OpenAI’s Chris Lehane confirmed the discussions have been underway “for several weeks” and said OpenAI doesn’t believe it needs an antitrust waiver. The group of companies compares the initiative to safety cooperation agreements between airlines. Sam Altman told OpenAI staff that the major labs would likely have to build the standards body themselves, without government backing, since “there was not yet enough industry consensus.” Reaction from AI executives outside the Big Three AI labs was both loud and pointed: Cohere’s Adam Gomez called the effort a “cartel” attempt to let “oligopolies” set rules for everyone else. FTC Chair Andrew Ferguson said he’d be “deeply suspicious” of any waiver request, calling it “moat digging.” Analyst Gil Luria of D.A. Davidson said Silicon Valley “builds a business really quickly exploiting regulatory ambiguity,” then “pulls the ladder” once big enough. A standards body sponsored by the best-funded of the big three AI labs is a proposal focused on how incumbents get regulated. Buyers evaluating vendors' commitments to any resulting standard should ask how the security standards were developed and who they benefit. Expect smaller labs and open-source players to keep pushing back until the governance terms are made public, and they are able to join the standards body and have a real say in what gets approved. Sources:The Information | Bloomberg | Cohere blog Anthropic’s latest threat intelligence report, Countering Misuse of AI: September 2026, shows what really bad actors tried to do with Claude. It’s not pretty: Militants in Yemen tried to use Claude’s coding tools to develop guidance software for rockets and missiles Iran-linked actors tried to target U.S. Navy ships State-linked groups in China ran distillation campaigns to copy Claude’s capabilities. Anthropic says its safeguards blocked most, but not all, of the activity. The report landed the same week Anthropic CEO Dario Amodei penned an essay arguing that the AI industry should slow the pace of AI model development. substackcdn.com The most alarming cases of attempted, but blocked, Claude-based malfeasance: Yemen-based militants tried to use Claude’s coding tools to replace human engineers on guidance, navigation, and control software, “hiding their goals” and splitting requests across sessions “so no single session revealed their full intent.” A test-fired guided rocket failed, and they returned to their workstations within hours to debug their code. Separately, Iran-affiliated individuals tried to use the model to target US Navy warships. A Mali-linked consultant attempted to build a domestic surveillance tool. Anthropic also detailed distillation campaigns from Alibaba, Moonshot AI, and DeepSeek, including one where Moonshot secretly routed nearly 300,000 queries through Claude via 5,380 fraudulent accounts.