TECH
Trump administration orders Anthropic to halt Fable and Mythos AI models
Anthropic disabled access to its Fable 5 and Mythos 5 models after receiving a U.S. Commerce Department directive that placed the models under export controls. Ars Technica reported the move was prompted by concerns that a narrow jailbreak could let the models produce cybersecurity, chemistry and biology information that might pose a national‑security threat. The Verge added that Amazon’s security research, shared with the White House, helped trigger the export‑control order and noted the action follows a broader conflict between the Trump administration and Anthropic over AI use policies.
SpaceX IPO values firm at $1.8 trillion as AI focus grows
SpaceX went public on the NASDAQ at $135 per share, ending the day near $161 and giving the company a market value of about $1.8 trillion, which makes its founder Elon Musk a trillionaire on paper. According to Ars Technica, the firm’s valuation is driven mainly by its planned AI services and orbital data‑center business, which it sees as accounting for the bulk of its future profit rather than its space‑flight or Starlink operations. The article notes that while NASA contracts such as the Artemis human‑landing system remain important, they are now dwarfed by multi‑billion‑dollar AI compute deals with firms like Anthropic and Google, signaling a shift toward profit‑focused AI work.
OpenAI under investigation by multiple state attorneys general
A coalition of state attorneys general, including New York, has issued a subpoena to OpenAI seeking documents on advertising, user engagement, data handling, and protections for minors and seniors. The probe comes amid other legal challenges, such as a lawsuit from Florida’s attorney general accusing the company of ignoring safety warnings for children and a recent trial involving co‑founder Elon Musk. OpenAI says it takes the concerns seriously, highlights safeguards for minors, and intends to work constructively with regulators.
PeopleSoft zero‑day exploited by ShinyHunters steals gigabytes of data
Ars Technica reported that a critical SSRF vulnerability (CVE‑2026‑35273) in Oracle’s PeopleSoft was exploited for two weeks before Oracle flagged it. The ransomware group ShinyHunters used the flaw to breach roughly 100 organizations, primarily in higher education, stealing tens of gigabytes of data and issuing extortion demands. Oracle has issued a stopgap mitigation but has not yet released a full patch, while Mandiant and Rapid7 have published indicators of compromise and remediation guidance.
KPMG withdraws AI report after hallucinations expose false usage claims
KPMG has pulled its October 2025 report “Redefining excellence in the age of agentic AI” after GPTZero flagged numerous hallucinated statements about AI use at several organizations. The UK’s NHS, UBS, Swiss Federal Railways and Transport for London said the report’s claims were inaccurate or misleading, prompting KPMG to remove the document while it investigates. The firm reiterated its AI guidelines requiring human oversight. The incident follows a similar withdrawal by EY of a loyalty‑rewards report containing fabricated footnotes.
Ukraine tests fully autonomous drones that killed Russian soldiers in one trial
Ars Technica reported that a Ukrainian drone maker claimed a one‑time field test used quadcopter drones programmed to enter a front‑line area and, in “Terminator mode,” autonomously seek and strike any target, killing a couple of Russian soldiers. Ukrainian officials, cited by New Scientist, say the military bans fully autonomous lethal decisions and relies on semi‑autonomous systems with human oversight to comply with international humanitarian law. Analysts note, per a CSIS report, that while fully autonomous weapons are still rare, both Ukraine and Russia are increasingly adding AI for navigation and limited target recognition to a growing array of combat drones.
Protests delay $130 billion in data center projects this year
NBC News citing Data Center Watch reports that protests have blocked or delayed 75 data‑center projects nationwide, representing about $130 billion in value during the first quarter of 2026, the highest three‑month total since tracking began in 2023. The coverage notes that researchers and sociologist Tressie McMillan Cottom, writing for the New York Times, say the opposition has become a coordinated “playbook” that is reshaping elections and prompting bipartisan political attention. Industry advocates, highlighted in The Atlantic and The Wall Street Journal, argue the sites generate significant tax revenue and jobs, but critics point to environmental‑review gaps and a growing legislative push for stricter oversight.
FISA section 702 expires but surveillance persists under existing certifications
Section 702 of the Foreign Intelligence Surveillance Act officially expired at midnight, but the current year‑long certification issued by the FISA Court will keep the program running until March 2027. Legal analysts cited by Ars Technica, including the Brennan Center and the Cato Institute, note that the law’s sunset clause was designed to allow continued data collection under those certifications, countering claims that surveillance will “go dark.” The Electronic Frontier Foundation adds that even after the certifications lapse, the government could rely on other authorities such as Executive Order 12333 to maintain overseas signals intelligence.
Jeff Bezos' Prometheus startup raises $12 billion to advance physical AI
Ars Technica reports that Bezos and co‑founder Vik Bajaj have secured a $12 billion funding round, valuing Prometheus at $41 billion, to build compute‑intensive “artificial general engineer” tools for physical AI. The company says the aim is to speed invention cycles and generate civilizational wealth, while a separate New York Times report notes plans for a $100 billion investment fund that could back ventures such as Blue Origin. CNBC is cited describing the need for massive compute resources to create the data that will power the effort.
Meta employees push back on Zuckerberg’s planned AI hackathon
Meta announced a companywide AI hackathon for July 14‑16, presenting it as a morale‑building effort. Employees, still coping with recent layoffs and increased workloads, voiced frustration on internal forums, saying they lack time and incentive to participate and questioning whether hackathon work will count toward performance reviews. WIRED reports the backlash included angry comments and sarcastic memes, while Meta declined to comment on the internal dissent.