September 23, 2026
Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.
AI
Anthropic launches Claude Opus 5.5
Anthropic has released Claude Opus 5.5, a model built for long-running agentic coding and knowledge work with a 1 million token context window by default, 128k max output tokens, and always-on adaptive thinking, priced at $4/$20 per million tokens. Lovable has already adopted it for faster builds, and early benchmarks and HN discussion point to strong performance and price positioning.
OpenAI launches GPT-6 Sol and Luna amid a new price war
OpenAI released GPT-6 Sol and Luna the same week as Anthropic's Claude Opus 5.5, triggering what Simon Willison calls a new price war among frontier model vendors. The llm command-line tool already added support for both new models.
OpenAI's GPT-6 Astra breaks an Enigma message unsolved since 2005
OpenAI's GPT-6 Astra has reportedly cracked a World War II-era Enigma-encrypted message that had resisted decryption since 2005. The break, documented by the cryptography community, has sparked debate over how far language models now reach into classical cryptanalysis.
Pentagon says overreliance on AI contributed to missile strike on Iranian school
The Pentagon says overreliance on AI-based decision tools contributed to a missile strike that hit a school in Iran, according to a Bloomberg investigation. The story has fueled debate over accountability and oversight when AI is used in military decision-making.
TypeSafe AI launches Jev, a new 'System One' decision-model category
TypeSafe AI has launched Jev, its first example of what it calls 'System One' models - Simon Willison prefers the term decision models. Willison has already built an LLM plugin for the model, and commentators are debating whether OpenAI will quickly fast-follow with a competing offering.
NVIDIA enables private AI inference with confidential computing
NVIDIA describes how confidential computing can deliver high-performance production AI inference without exposing sensitive information or proprietary model context, as LLM inference increasingly processes regulated and sensitive data.
Security
ShinyHunters claims it hacked the FBI via an Oracle PeopleSoft zero-day
The ShinyHunters extortion gang claims it breached FBI systems by exploiting a zero-day vulnerability in Oracle PeopleSoft and says it obtained data on all FBI employees. Both BleepingComputer and 404 Media are covering the claim, which the FBI has not yet confirmed.
Chinese hackers exploited WordPress and Zyxel flaws to steal government data
A Chinese-speaking threat actor exploited vulnerabilities in Zyxel GS1900 switches and WordPress to steal sensitive data from nearly 1,000 devices and more than 18,500 backend database records, according to BleepingComputer.
Microsoft disrupts EvilTokens phishing service after 12,000 accounts compromised
Microsoft's Digital Crimes Unit led an effort that disrupted EvilTokens, a phishing-as-a-service platform that compromised more than 12,000 Microsoft accounts across over 10,000 organizations.
New ClosedQuorum malware uses AI models to steer attacks
A new Windows malware strain called ClosedQuorum uses Google Gemini, DeepSeek, Qwen, and Mistral AI models to autonomously decide what actions to take during the post-compromise stages of an attack, according to BleepingComputer.
Dev
Cloudflare's Python Workers reach general availability
After a two-year preview, Cloudflare's support for running Python code on its server-side Workers platform is now stable and generally available, Simon Willison reports.
Other
Apple adds persistent ads to iOS, frustrating users
Apple has started showing persistent ads within iOS, a change users say is making the experience increasingly annoying. The story drew heavy engagement on Hacker News amid a broader debate about advertising creeping into premium products.