September 11, 2026
Anthropic report details alleged Claude misuse from cyberattacks to weapons research
IDScan says customer ID data may have been accessed or copied, while Apple leaves standard iPhone 18 out of its fall showcase.
Anthropic report details alleged Claude misuse from cyberattacks to weapons research
Anthropic has published a report on cases in which it says Claude was misused across cyber operations, influence campaigns, surveillance, weapons research and model distillation.
- Anthropic says it identified and disrupted the activity between December 2025 and August 2026. The cases include familiar cyberattacks accelerated by AI, propaganda production, surveillance and potential military work.
- The company says it investigated five potentially dangerous biological-research cases, but did not conclude that the scientists intended harm.
- Anthropic also alleges that Alibaba, Moonshot AI, DeepSeek and Zhipu used fraudulent accounts or relayed requests to extract Claude outputs for training.
IDScan says customer ID data may have been accessed or copied
IDScan has acknowledged a security incident involving customer data in its cloud platform.
- The company says an unauthorized third party may have accessed or copied names and government-issued ID numbers.
- The notice followed Brian Krebs’ report linking IDScan to a dark-web database advertised as containing more than 150 million U.S. and Canadian driver’s-license records. IDScan has not disclosed how many people were affected.
Apple leaves standard iPhone 18 out of its fall showcase
Apple unveiled its first foldable phone, the iPhone Duo, alongside the iPhone 18 Pro and Pro Max at John Ternus’ first launch showcase as CEO.
- A standard iPhone 18 was absent, breaking Apple’s recent pattern of introducing standard and Pro models together.
- Analysts cited by CNBC expect the standard model in spring, though Apple has not announced a date. They warn price-sensitive buyers could wait for lower-priced models.
OpenAI’s Agents API offers managed access to the Codex harness
OpenAI has documented a managed Agents API that gives applications access to the Codex harness.
- The company says it handles sessions, orchestration, context compaction and recovery, while developers provide tools and choose an execution environment.
- OpenAI says agents can use sandboxes to run code and edit files. Its documentation lists separate standard charges for model use, OpenAI tools and hosted sandboxes.
Microsoft plans to more than triple data-center capacity by 2032
Microsoft plans to more than triple its global data-center capacity by 2032, Bloomberg reports, citing people familiar with the plans.
- Capacity would rise from about 12 gigawatts to more than 38 gigawatts, with about one-third centered on AI-specific chips.
- Bloomberg says computing shortages have forced Microsoft to turn away some AI and cloud business.
Newsom signs social media and chatbot safeguards for minors
Gov. Gavin Newsom signed 13 child online-safety bills, CalMatters reports.
- Among the measures described, social platforms must disable infinite scroll and autoplay for users under 16 or keep them off their services.
- Chatbot operators must impose time limits, provide mental-health resources and alert parents to detected self-harm threats, with potential liability for failing to act. Effective dates were not reported.
In Brief
Pentagon discusses roughly $5 billion Fluidstack loan
The Pentagon is discussing a roughly $5 billion loan to AI cloud startup Fluidstack, The Wall Street Journal reports. If completed, it would be the Office of Strategic Capital’s largest loan by far.
OpenAI considers slowing cutting-edge AI development
Sam Altman told employees OpenAI could slow the pace of cutting-edge AI development, potentially alongside other labs, Bloomberg reports, citing unnamed sources. He acknowledged that some labs might not participate.
DeepSeek rolls out V4.1-Flash model
DeepSeek launched V4.1-Flash, which it calls the smallest model in its new architecture family. Reuters reports that it uses a Causal Encoder-Decoder architecture, with 552 billion backbone parameters and a 1 million-token context window.
Amazon DSP adds ChatGPT ad inventory in U.S. pilot
Amazon and OpenAI are piloting ChatGPT ad inventory for select U.S. advertisers using Amazon DSP. Amazon says its team will manage campaigns, while OpenAI’s system controls which ads are delivered.
GreyNoise links AI-assisted PaperCut attack to 395 organizations
GreyNoise says an attacker used hundreds of AI agents to exploit two PaperCut NG/MF flaws, compromising at least 440 instances across 395 organizations. Administrators were advised to apply PaperCut’s emergency updates immediately, BleepingComputer reports.
Oracle’s cloud infrastructure revenue jumps 121% as spending climbs
Oracle’s cloud infrastructure revenue jumped 121% to $7.4 billion, helping total revenue rise 30% to $19.35 billion. Quarterly capital spending climbed to $28.50 billion from $8.50 billion a year earlier.
Ant International, Visa and Mastercard plan common AI payment standard
Ant International has enlisted Visa and Mastercard to develop common standards for AI-agent payments, CNBC reports. Ant says the proposed framework would verify agents, monitor their behavior and work across card and e-wallet systems.
