Daily · AI Price War and Geopolitical Shifts · September 23, 2026
Key points
- Xi Jinping arrives in Washington for a state visit, skipping the UN General Assembly to meet Donald Trump on trade and AI.
- OpenAI and Anthropic launch new models with significant price cuts, intensifying the frontier AI cost competition.
- ShinyHunters claims a breach of FBI systems, alleging the theft of personnel data via an Oracle PeopleSoft zero-day.
- Texas Governor Greg Abbott orders a halt to all data center permits pending an audit of the existing waitlist.
Xi-Trump Summit and AI Diplomacy
Chinese President Xi Jinping has arrived in Washington for a state visit from September 23 to 25, bypassing the UN General Assembly in New York. This is his first official state visit to the US since 2015 and follows an invitation extended during Donald Trump's May trip to Beijing. The itinerary includes a personal reception at Joint Base Andrews, a White House arrival ceremony, and a state dinner expected to be attended by tech leaders including Mark Zuckerberg, Jeff Bezos, and Elon Musk.
Trade and artificial intelligence are the central topics. US officials indicate the sides may ease tariffs on non-strategic goods and discuss rare earths, while China's Foreign Minister Wang Yi has emphasized that Taiwan and human rights remain "red lines." The summit occurs against a backdrop of a trade truce set to expire in November and ongoing tensions over AI export controls.
In related diplomatic moves, South Korean President Lee Jae Myung pledged to double the country's semiconductor production capacity within five years and build AI supply chains resilient to external shocks. He called for fresh US investment in these "three megaprojects," aiming to secure strategic autonomy in chip manufacturing.
Frontier Model Price War
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, cutting prices by 50% compared to their GPT-5.6 predecessors. GPT-6 Sol is priced at $2 per million input tokens and $10 for output, while GPT-6 Luna costs $0.10 and $0.50 respectively. OpenAI reports that GPT-6 Sol achieves performance close to its flagship GPT-6 Astra on coding benchmarks at a fraction of the cost, with a 90% discount on cached input reads.
Anthropic launched Claude Opus 5.5 on the same day, reducing prices by 20% from its predecessor to $4 per million input tokens and $20 for output. Anthropic claims Opus 5.5 performs at the level of its top-tier Fable 5.1 model on most work tasks while generating output 30% faster. The simultaneous releases signal a shift toward making high-performance AI more affordable, with both companies citing efficiency gains in their underlying architectures.
This price competition coincides with a public call for caution. Anthropic CEO Dario Amodei urged tech leaders to slow AI development following a post by a former employee warning of existential risks. OpenAI CEO Sam Altman and Elon Musk endorsed the call, though both companies continued to release new models.
Cybersecurity Breaches and Threats
The hacker group ShinyHunters claimed a breach of the FBI on September 22, stating they used a zero-day exploit in Oracle PeopleSoft to access AWS GovCloud servers. The group alleges it exfiltrated between two and three terabytes of data, including personal information for approximately 5,000 FBI employees. Reuters partially verified some of the stolen personnel data, and an FBI spokesperson confirmed the agency is investigating unauthorized activity affecting its job portal.
Microsoft led an industry-wide disruption of EvilTokens, a subscription-based scam platform that used an AI chatbot to compromise 12,000 Microsoft accounts across 10,000 organizations. The platform abused legitimate OAuth device code authentication to gain access. Microsoft seized 50 websites and 150 domains associated with the service, and the UK Metropolitan Police arrested two men in connection with the operation.
Additionally, a new Windows malware dubbed CLOSEDQUORUM has been identified as the first publicly documented implant to use large language models for command-and-control decisions. This development complicates endpoint detection, as post-compromise actions are no longer fixed scripts but are selected autonomously by AI models.
Infrastructure and Regulatory Moves
Texas Governor Greg Abbott ordered the state's environmental regulator to halt all data center-related permits while an audit of the existing waitlist is completed. This move follows Abbott's August plan to audit proposed data centers seeking to connect to the Texas power grid. The decision reflects growing political opposition to data center expansion, with 57% of Texas voters polled by the Texas Politics Project opposing a data center in their community.
In Europe, the European Supervisory Authorities (EBA, EIOPA, ESMA) identified external dependencies, emerging technologies, and private credit as key vulnerabilities for the EU financial system. The Autumn 2026 risk update warned that reliance on non-EU ICT providers could amplify geopolitical shocks and highlighted growing cyber risks linked to AI models.
SoftBank is set to price a mega junk-bond deal expected to reach about $11 billion across two currencies to finance its AI build-out. This marks the AI infrastructure sector's crossover into high-yield debt, signaling rising capital costs for technology companies.
AI Governance and Legal Precedents
British Columbia filed a lawsuit against OpenAI and Sam Altman, demanding they pay for the rebuilding of Tumbler Ridge Secondary School and cover emergency costs following a mass shooting where an 18-year-old used ChatGPT to plan violence. The province also seeks injunctive relief requiring ChatGPT to automatically terminate violent conversations.
Norway and Finland called for human control of AI at the UN, gathering signatures from nearly two dozen world leaders including Germany, the United Arab Emirates, and Canada. This push came after US President Trump rejected global AI oversight and suggested renaming AI "superintelligence." French President Emmanuel Macron also called for a joint open-source frontier AI model independent of the US and China.
In the UK, Prime Minister Andy Burnham announced the establishment of a National Centre for Information Defence and a new AI defence partnership with the United States to protect critical national infrastructure. He also proposed working towards a single set of global principles and standards for AI transparency.
Our read
The simultaneous arrival of Xi in Washington and the aggressive price cuts by OpenAI and Anthropic suggest that economic competition is currently taking precedence over ideological friction in US-China relations. For technology leaders, the 50% drop in frontier model costs significantly improves the unit economics of AI-driven products, potentially accelerating adoption in regulated sectors where cost efficiency is a key constraint. However, the Texas permit halt and the FBI breach highlight that infrastructure expansion and digital sovereignty remain politically volatile; companies should anticipate stricter local oversight and enhanced security requirements for cloud and data center deployments in the near term.
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
- Chinese leader Xi Jinping will skip the UN General Assembly and instead head to Washington for a high-level summit with US President Donald Trump on Wednesday — asia.nikkei.com, thediplomat.com, en.mercopress.com (+27)
- Anthropic released Claude Opus 5.5 and OpenAI released GPT-6 Sol and GPT-6 Luna on the same day, with GPT-6 Sol and Luna priced at half the cost of their GPT-5.6 equivalents and Opus 5.5 cut 20% from the Opus 5.0 price, intensifying a new AI model price war — simonwillison.net, thenewstack.io, venturebeat.com (+11)
- Anthropic introduced Claude Opus 5.5, the first model in the Claude 5.5 family, designed to complete full software development lifecycle tasks — thenewstack.io, venturebeat.com, artificialanalysis.ai (+3)
- ShinyHunters claimed a breach of the FBI, sharing a screenshot of a vandalized FBI job site and a sample of allegedly stolen personnel data — 404media.co, dailymaverick.co.za, theregister.com (+2)
- The Register reported that Windows malware dubbed CLOSEDQUORUM uses AI models to autonomously select post-compromise actions, described as the first publicly documented Windows implant to use LLMs for command-and-control — theregister.com (+3)
- Meta hot-fixed a zero-day vulnerability in its Muse Mac app within a day of disclosure, but the incident exposed that enterprise security teams lack central visibility into what the AI agent can access on connected services — venturebeat.com, wired.com, cnbc.com (+1)
- South Korean President Lee Jae Myung unveiled a plan to double the country's semiconductor production capacity within five years and build AI supply chains free from external shocks, calling for fresh US investments — en.yna.co.kr (+2)
- China's rare-earth export controls are creating supply and planning risks for Swiss industry, with a decision expected by November 2026 on whether to apply a stricter regime that includes extraterritorial re-export authorisation and technology export controls — nzz.ch, bbc.co.uk, thediplomat.com
- JetBrains formally unveils JetBrains Air as an open system of products for agentic software development — thenewstack.io
- Microsoft led an industry-wide disruption of EvilTokens, a subscription-based scam platform that used an AI chatbot to compromise 12,000 Microsoft accounts — arstechnica.com
- DeepSeek publishes DSec, a production sandbox infrastructure for agentic training at scale — arxiv.org
- Anthropic CEO Dario Amodei urged other tech leaders to slow AI development after a former employee's post, with OpenAI CEO Sam Altman and Elon Musk endorsing the call — cnbc.com
- British Columbia filed a lawsuit against OpenAI and Sam Altman demanding they pay for rebuilding Tumbler Ridge Secondary School and cover emergency costs after a ChatGPT-assisted mass shooting, and seeking injunctive relief to require ChatGPT to automatically terminate violent conversations — arstechnica.com
- SoftBank is set to price a mega junk-bond deal expected to reach about $11 billion across two currencies to finance its AI build-out — marketwatch.com
- The European Supervisory Authorities (EBA, EIOPA, ESMA) identified external dependencies, emerging technologies, and private credit as key vulnerabilities for the EU financial system in their Autumn 2026 risk update — eba.europa.eu
- Texas Governor Greg Abbott ordered the state's environmental regulator to halt all data center-related permits while an audit of the existing waitlist is completed — cnbc.com
- New Chinese entry-exit rules took effect last week, giving authorities the power to stop engineers, founders, and other specialists from leaving the country if their expertise in batteries, rare earths, or AI is judged a threat to 'industrial and technological security.' — dw.com
- French President Emmanuel Macron delivered his final UN General Assembly speech, calling for a moratorium on strikes on civilian and energy infrastructure in the Russia-Ukraine war and for a joint open-source frontier AI model independent of China and the US — france24.com, news.un.org, timesofisrael.com (+3)
- EQT bought a majority share in Swiss cybersecurity company Acronis at an equivalent of a $3.5 billion or higher valuation for the entire firm — theregister.com (+4)
- Shopify acquired the open-source CSS framework Tailwind, giving it a stable long-term home as vibe coding eroded the web development platform's bottom line — theregister.com (+3)
- Switzerland is testing a free-and-open-source-software escape route from Microsoft 365 with its own FOSS push — theregister.com (+2)
- Apple agreed to pay $250 million in a class action settlement over claims it overhyped Apple Intelligence and the delayed Siri AI overhaul — wired.com, marketwatch.com, rbc.ru
- Cognex acquired Intel's robotics spinoff RealSense, developer of 3D cameras and AI vision technology, for over $600 million in its first Israel acquisition — timesofisrael.com (+1)
- Ofcom launched an investigation into Pornhub owner Aylo over whether the site's new age checks are effectively stopping children from accessing adult content — bbc.co.uk, politico.eu
- NEC will handle design and construction for Meta Platforms' Petal undersea cable project connecting the U.S. and France, targeting the world's first petabit-class underwater connection — asia.nikkei.com