Daily · AI Misuse, Finance Launches, and Chip IPOs · September 11, 2026

Key points

Anthropic report details state-sponsored misuse of Claude

Anthropic published a threat intelligence report covering December 2025 to August 2026, detailing cases where foreign actors circumvented restrictions to use Claude for illicit purposes. The company stated it disrupted every identified case and strengthened safeguards on Sonnet, Opus, and Haiku models.

The report identified five cases involving biological weapons research, including a scientist using Claude to draft a grant application for gain-of-function research on a mosquito-borne virus and another researching adaptations of highly pathogenic avian influenza. In the military hardware sector, Anthropic stopped a Yemeni group from developing guidance software for guided rockets and ballistic missiles, and identified a Russian freelancer attempting to build kamikaze drones. Additionally, state actors from Iran and China used the models to analyze social media for dissidents, while a consultant in Mali used Claude to code a mass-interception platform for mobile operators.

Bad actors accessed the models via VPNs and fraudulent accounts, often facilitated by intermediaries helping customers bypass geographic restrictions. Anthropic noted that none of these incidents involved its most powerful model, Claude Mythos, which is restricted to trusted partners. The company shared intelligence with governments and industry groups, emphasizing that misuse has expanded beyond Chinese spies to include groups like ShinyHunters and Russian freelancers.

OpenAI launches ChatGPT for Financial Services

OpenAI launched ChatGPT for Financial Services on September 10, an industry-specific product built on GPT-6 Astra. Developed with Morgan Stanley and Evercore, the platform bundles premium financial data from Daloopa, PitchBook, and LSEG News, eliminating the need for separate data contracts or connector setup.

In OpenAI's internal evaluation on OfficeQA Pro, GPT-6 Astra scored 69.9% correctness, outperforming GPT-5.6 Sol (60.2%) and Claude Fable 5.1 (62.4%). The product includes enterprise governance controls and firm-controlled templates, with shared sign-in integrations in development for S&P Capital IQ, MSCI, and Moody's. The wider ecosystem supports more than 50 integrations, including Datasite and Box.

Pricing and minimum seat requirements have not been publicly disclosed. The launch positions the product as a direct competitor to Microsoft Copilot and incumbent data platforms, with governance and provenance as key differentiators for financial institutions.

Enflame IPO surges as Chinese chip demand grows

Enflame, a Chinese AI chipmaker backed by Tencent, surged 188% in its Shanghai stock-market debut on September 11. The company raised $910 million in the IPO, which was priced at a lower valuation compared to peers but saw orders for over 6,000 times the shares available.

Enflame is the last of China's 'four little dragons' of AI chipmaking to go public, following MetaX (700% first-day gain), Moore Threads (400%), and Biren (76%). The company reported revenue of 990 million yuan ($147 million) in 2025, up from 722 million yuan the previous year. International chipmakers led by Nvidia accounted for nearly 60% of China's AI accelerator market in 2025, but U.S. export controls and Beijing's push for self-sufficiency are driving domestic procurement.

Goldman Sachs expects China's semiconductor capital spending to reach $82 billion by 2030. The strong debut signals continued investor appetite for Nvidia alternatives and may accelerate procurement shifts in the region.

SpaceX secures $1.1B monthly AI compute deal

SpaceX CFO Bret Johnsen revealed at a Goldman Sachs conference that the company secured a new AI computing-power deal worth $1.1 billion per month starting in December. This translates to approximately $13 billion in annualized revenue, bringing SpaceX closer to its goal of a $100 billion annualized revenue run rate by the end of 2026.

The new contract joins existing deals with Anthropic ($1.25 billion per month until May 2029) and Google ($920 million per month through June 2029). Johnsen noted that almost all of SpaceX's computing deals allow both sides to end the contract prematurely after just a few months, creating revenue-visibility risk for firms modeling the company's growth. Over $36 trillion is expected to be spent on computing capacity by 2050.

Apple debuts iPhone Duo and raises device prices

Apple introduced the iPhone Duo, its first foldable phone, at a product event led by new CEO John Ternus. The device is priced at $1,999, though some reports cite a starting price of $1,799 for the 'iPhone Fold.' Apple also raised prices on newer and earlier-generation phones by $100, bucking its trend of discounting older models.

The iPhone 18 Pro starts at $1,199 and the Pro Max at $1,299. Carriers are offering significant trade-in credits: T-Mobile offers up to $1,200 off, while AT&T and Verizon have similar promotions. Apple Upgrade allows Duo buyers to lease for $57.99 per month over two years. Shares of Apple rose 3.6% on the news.

Analysts noted that the price hikes are below the $150 to $200 increases seen on iPads and MacBooks this summer. JPMorgan analysts said favorable pricing moves and AI features will drive robust iPhone revenue growth, while Evercore ISI described carrier promotions as 'targeted escalation' rather than a price war.

Regulatory and security developments in AI and tech

California Governor Gavin Newsom signed 13 youth online safety laws, including SB 1119, which mandates age verification for AI chatbots and limits advertising to minors. The law was backed by OpenAI after clarifications on memory retention were added. AB 1709 bans personalized feeds that maximize screen time for children under 16.

In cybersecurity, Microsoft patches failed to fix a vulnerability in on-premises SharePoint, which is now under active zero-day attack. Separately, ClickFix attacks using fake CAPTCHA overlays have become mainstream, infecting PCs and Macs at scale by exploiting user fatigue with interstitial prompts.

Google signed an agreement with the Pentagon allowing use of its AI models for classified operations, despite a petition from over 600 employees opposing military access. The deal requires Google to assist in setting up protective measures but does not allow limiting use within lawful operations. In contrast, Anthropic refused to meet Pentagon conditions and was subsequently banned from government use by President Donald Trump.

Our read

The convergence of Anthropic's misuse report and the Google-Pentagon deal highlights a widening gap between AI safety commitments and state-level security demands. For decision-makers, this suggests that 'responsible' AI deployment is increasingly defined by geopolitical alignment rather than technical safeguards alone. The launch of ChatGPT for Financial Services and Enflame's IPO debut indicate that capital is flowing toward both specialized vertical applications and sovereign hardware alternatives, forcing a re-evaluation of vendor lock-in strategies. Finally, the active zero-day in SharePoint and the mainstreaming of ClickFix attacks underscore that traditional perimeter defenses are failing against social-engineering vectors, necessitating immediate updates to endpoint detection and user-awareness protocols.

This material was produced automatically by a large-language-model system from the public sources listed below; it is AI-generated content and may contain inaccuracies — verify facts against the original sources.

Sources