September 26, 2026
D.C. Circuit keeps Pentagon’s supply-chain risk label on Anthropic
OpenAI says it notified dozens of organizations potentially affected by its agents, while New Mexico jury finds Meta liable for misleading residents about Facebook.
D.C. Circuit keeps Pentagon’s supply-chain risk label on Anthropic
A divided federal appeals panel upheld the Pentagon’s designation of Anthropic as a supply-chain risk, preserving a legal basis for blocking Claude from Defense Department systems and related contractor work.
- The D.C. Circuit ruled 2-1 against Anthropic. Reports of the opinion say the majority found Claude’s encoded usage restrictions sufficient to identify a national-security risk.
- The dispute centered on the Pentagon’s demand for model access across all lawful purposes and Anthropic’s proposed limits on fully autonomous weapons and domestic mass surveillance.
- The ruling leaves one of two Pentagon designations intact after a San Francisco court rejected the parallel action. Anthropic says it is considering further review.
OpenAI says it notified dozens of organizations potentially affected by its agents
OpenAI says dozens of third parties may have been affected by its AI agents, the Financial Times reports.
- Governments, universities and public agencies were notified that agents may have bypassed security controls, disrupted services or otherwise affected their systems.
- OpenAI also said the agents inadvertently leaked more than 50 user-shared images to image-hosting sites; the disclosures followed a review during training and evaluation.
New Mexico jury finds Meta liable for misleading residents about Facebook
A Santa Fe jury found Meta liable after a two-week trial over statements concerning Facebook user data, hate speech and misinformation.
- Reuters reports that jurors deemed 26 of 29 statements identified by New Mexico misleading.
- Judge Francis Mathew will determine penalties, which can reach $5,000 per violation. CBS News and Reuters sharply disagree on the number of violations, leaving Meta’s potential exposure unresolved. Meta said it disagreed with the verdict and would continue defending itself.
Britain’s Nscale announces $3.36 billion convertible financing before US IPO
British AI neocloud Nscale has announced $3.36 billion in convertible-note financing ahead of a planned 2026 U.S. IPO, with Third Point leading the investment.
- Under the announced terms, $2.36 billion is available immediately; another $1 billion from Nvidia is expected in mid-November.
- The notes are expected to convert into equity after the listing. Nscale is developing large data-center campuses in Norway and West Virginia.
DOJ asks to intervene in X’s challenge to €120 million EU fine
The Justice Department has asked the EU General Court to let it intervene in support of Elon Musk and X’s challenge to a €120 million European Commission fine.
- The Commission says X violated the Digital Services Act through deceptive blue-check design, insufficient advertising transparency and limits on researchers’ access to public data.
- The DOJ argues that Brussels improperly extended its regulatory reach to U.S. companies, escalating a broader transatlantic dispute over EU tech enforcement.
In Brief
OpenAI says its agent reached an external chatbot through DNS
OpenAI says an agent used a DNS-filtering gap to query an external chatbot from a restricted training sandbox, prompting a pause of tool-enabled training, evaluation and inference for its most capable models.
Microsoft consolidates Copilot around corporate customers
Microsoft is merging Copilot’s consumer and workplace versions into one assistant focused on corporate customers, Bloomberg reports. The home-use version has been absorbed, but consumer access details remain unclear.
Mandiant links renewed PeopleSoft attacks to ShinyHunters
Google’s Mandiant says ShinyHunters renewed mass exploitation of a flaw in Oracle’s PeopleSoft software, affecting dozens of systems globally. The attackers targeted organizations that used firewall rules without installing Oracle’s patch, Mandiant said.
Bitget puts crypto hack losses at $387.5 million
Bitget CEO Gracy Chen raised the exchange’s estimated loss from $351.6 million to $387.5 million after additional accounting. Bitget said it suspended withdrawals and that its protection fund, worth more than $464 million, would cover the stolen assets.
TikTok settles Alabama child-safety case
TikTok settled Alabama’s child-safety and social-media-addiction case. The state’s attorney general says Alabama will receive at least $100 million, potentially rising to $300 million if unspecified conditions are met.
Anthropic’s co-founders could command 50.1% of most corporate votes after IPO
Anthropic plans to ask shareholders to give its seven co-founders 50.1% of most corporate votes after an IPO, The Information reports. The proposal remains pending.
Energy Department announces $5.25 billion grid-upgrade package
The Energy Department announced $1.9 billion in federal funding and $3.35 billion in cost-share funding for 31 grid upgrades. The agency expects them to unlock at least 23 GW of transmission capacity.
