Rival AI Labs Team Up on Safety as FTC Chief Smells an Antitrust Play

AI & Machine Learning

OpenAI’s global policy chief Chris Lehane confirmed on September 15 that OpenAI, Anthropic, and Google DeepMind have spent weeks quietly coordinating on AI safety, following Anthropic CEO Dario Amodei’s public call for the industry to collectively slow frontier development. Lehane said the three labs don’t believe they need a government antitrust waiver to work together, and that discussions include support for provisions in the proposed FRONTIER Act requiring independent verification organizations inside frontier labs. The talks mark an unusual truce between fierce commercial rivals, driven by shared concern that AI agents could pose catastrophic risks if deployed without common safeguards. The move drew immediate skepticism from the Trump administration, which worries that a coordinated slowdown could cede ground to China in the AI race. Source: TechCrunch Verified: True

Speaking at the Semafor Future of Health Forum on September 15, Isomorphic Labs’ head of drug discovery Chris Butler said the Google DeepMind spinout has no plans to slow down despite the industry-wide debate over pacing frontier AI. “We’re able to still progress regardless of what decisions are made elsewhere,” Butler said, noting that Isomorphic’s models are “locked down in-house” and insulated from decisions made by OpenAI, Anthropic, or Google. The company, which recently raised $2.1 billion and works with Novartis, Eli Lilly, and Johnson & Johnson, is pushing ahead on oncology and immunology programs as it tries to compress the traditional drug-discovery timeline. Butler’s comments illustrate a growing split in the AI industry between labs racing toward frontier capability and those, like Isomorphic, applying already-built models to narrower, high-value problems. Source: Semafor Verified: True

Consumer Hardware

UBTech’s UWorld brand began home deliveries of its U1 companion humanoid robots in China on September 16, fulfilling a backlog of more than 13,000 pre-orders that arrived after sales opened in June. The lineup spans three variants — Lite, Pro, and Ultra — priced from roughly $16,500 to $135,000, with the full-body Pro and Ultra models standing up to 1.83 meters tall, featuring medical-grade silicone skin, 88 joints for lifelike movement, and an “emotion-aware” AI model with locally encrypted memory. Industry watchers have framed the launch as UBTech’s entry into China’s “loneliness economy,” selling AI companionship to isolated buyers, though early adopters have reported gaps between the robots’ promotional demos and real-world performance. The rollout represents one of the largest consumer humanoid deployments to date, roughly ten times UBTech’s total 2025 humanoid sales. Source: Startup Fortune Verified: True

Meta is reportedly preparing a new line of smart glasses, internally called Luna, that ditch onboard cameras entirely in response to backlash over its existing camera-equipped Ray-Ban models being labeled “invasive emblems” of surveillance. Luna is said to rely on six built-in microphones and a side button for hands-free access to Meta AI and the company’s Muse AI agent, letting the slimmer temple design forgo a lens-mounted camera altogether. The glasses could debut as soon as Meta Connect on September 23-24 in Menlo Park, with shipments potentially starting in October, expected in Clubmaster and Burbank styles. The camera-free pivot signals Meta is willing to trade a marquee feature for broader consumer trust as regulators and privacy advocates keep pressure on always-on recording wearables. Source: TechCrunch Verified: True

Cybersecurity

CenterPoint Energy confirmed on September 14 via an SEC filing that hackers exploited an unauthenticated, rate-limit-free API to extract customer data by cycling through account IDs, in a breach a threat actor first claimed on the dark web on September 1. Roughly 7.5 million raw records were reportedly stolen, including customer names, phone numbers, email and billing addresses, account numbers, billing amounts, and the last four digits of Social Security numbers, out of about 7 million customers the utility serves across Texas, Indiana, Minnesota, and Ohio. Five class-action lawsuits were filed against the company before it even made the breach official, alleging it failed to maintain basic data-security safeguards on the exposed endpoint. CenterPoint says electricity and gas service were not affected and that it has engaged third-party investigators and law enforcement. Source: Hoodline Verified: True

CISA added two actively exploited vulnerabilities to its Known Exploited Vulnerabilities catalog on September 16: CVE-2026-76460, an incorrect-use-of-privileged-APIs flaw in Cisco Identity Services Engine, and CVE-2026-87886, an incorrect default-permissions vulnerability in Acronis Backup. Both entries were added on evidence of active exploitation in the wild, and the advisory points Federal Civilian Executive Branch agencies to Binding Operational Directive 26-04, which requires rapid remediation of KEV-listed flaws on internet-facing assets. Cisco ISE bugs are a recurring target for attackers because the platform underpins network access control at large enterprises, while Acronis Backup’s exposure raises the stakes for organizations relying on it as a last line of defense against ransomware. The additions extend a busy month for CISA, which has already flagged more than a dozen newly exploited CVEs in the first two weeks of September. Source: CISA Verified: True

Enterprise Infrastructure

Workflow-orchestration platform Temporal closed a $550 million Series E on September 14 at a $12.55 billion valuation, in a round co-led by Lightspeed, Wellington Management, Goldman Sachs Alternatives’ Growth Equity arm, and Tiger Global, with T. Rowe Price and SV Angel also joining. Temporal’s “Durable Execution” technology lets developers write ordinary application code while the platform handles state persistence and automatic recovery across distributed systems, a capability CEO Samar Abbas says has become critical as AI agents take on tasks spanning days or weeks. The company reported its annualized revenue run rate is up more than 200% year over year, with more than 4,300 paying customers including OpenAI, Snap, and Nvidia, and said it processed 1.9 trillion billable actions in August alone. The raise underscores how reliability infrastructure, once a niche concern, has become a flashpoint of enterprise AI spending as agentic workloads scale. Source: Temporal Verified: True

Salesforce suffered a global outage on September 16, the second day of its Dreamforce conference, after an internal login service began stalling under load and consumed available server resources, disrupting access to parts of its cloud suite for customers across the US, UK, Germany, France, India, and Japan. The outage began around 8:30 a.m. UTC, and Salesforce validated a fix on a test system by 10:56 a.m. UTC before rolling it out across affected infrastructure. The timing was awkward for the company, which was simultaneously pitching new AI products and enterprise-grade reliability to major clients at Dreamforce; Salesforce nonetheless reported quarterly revenue of $11.35 billion, up 11% year over year, and raised its full-year guidance by $200 million. The incident adds to a string of high-profile cloud outages in 2026 that have renewed enterprise scrutiny of concentration risk in hyperscale SaaS platforms. Source: KuCoin News Verified: True

Policy & Regulation

Governor Gavin Newsom signed Senate Bill 1050 on September 16, requiring explicit disclosure on any video or audio advertisement that uses AI-generated performers to sell a product or service, with violating ads barred from continued use. The law, sponsored by SAG-AFTRA and authored by Senator Ashby, is aimed at protecting creative-industry workers from being displaced by synthetic performers without audiences knowing what they’re watching. “Californians deserve to know when the person selling them something isn’t a person at all,” Newsom said in signing the bill, which builds on the state’s 2024 digital-likeness protections and its broader AI Transparency Act watermarking requirements. The signing caps a busy legislative stretch for California AI law, following child-safety chatbot rules and an AI auditor registry signed earlier in September. Source: Office of Governor Gavin Newsom Verified: True

FTC Chairman Andrew Ferguson said on September 15 that AI companies simultaneously lobbying for new regulation and for antitrust exemptions should set off “all of my alarm bells,” speaking at a Georgetown University event days after Senators Jim Banks and Adam Schiff sought to attach a narrow AI antitrust exemption to the annual defense policy bill. Ferguson argued such requests risk creating “barriers to entry that will insulate their incumbency from challenge,” and said the pattern should make observers “deeply suspicious,” even as he acknowledged legitimate safety coordination shouldn’t require a waiver at all. His remarks came just as OpenAI, Anthropic, and Google DeepMind’s safety-coordination talks became public, adding a competition-law wrinkle to what the labs have framed as a purely safety-driven effort. The clash signals that the Trump-era FTC intends to scrutinize AI industry coordination on competitive grounds even as it otherwise favors a light regulatory touch. Source: InsideAI News Verified: True