OpenAI exploit chain raises alarms & NYT lawsuit reveals AI concerns - Hacker News (Sep 18, 2026)

Please support this podcast by checking out our sponsors: - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad - Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron - Lindy is your ultimate AI assistant that proactively manages your inbox - https://try.lindy.ai/tad Support The Automated Daily directly: Buy me a coffee: https://buymeacoffee.com/theautomateddaily Today's topics: OpenAI exploit chain raises alarms - Researchers say a libheif bug and an OpenAI SSO misconfiguration were chained to reach employee ChatGPT and Codex accounts. The story highlights AI security, account compromise, image parsing risk, and how offensive research is getting faster. NYT lawsuit reveals AI concerns - New unredacted filings in The New York Times lawsuit allege OpenAI and Microsoft internally recognized legal and business risks around training on news content. Key issues include copyright, fair use, publisher traffic loss, paywalls, and AI chatbot substitution. Tao reframes AI math risks - In the latest AI and mathematics debate, Terence Tao says the biggest danger may not be a flood of useless theorems, but mathematicians losing the deep problem-solving work that builds intuition. It is an important update on AI, research culture, and mathematical practice. Hister keeps personal search local - Hister is an open-source personal search engine that indexes web pages you have seen and local files you control, with privacy-first deployment on local or self-hosted systems. Keywords here are personal search, self-hosting, privacy, browser history, and local AI workflows. Wax motors power quiet automation - A wax motor uses expanding and contracting wax to turn heat into smooth linear motion, and it quietly shows up in appliances, HVAC systems, and other everyday machines. It matters as a simple, reliable example of mechanical automation without complex electronics. - Hister: A Private Personal Search Engine - Fujitsu Unveils MONAKA CPU and Sovereign AI Server - OpenAI Launches Astra for Law - Bend: A Fast Language That Uses Proofs to Block AI Mistakes - PrismML Releases Bonsai 2 27B, a 9x Smaller Near-Lossless AI Model - Wax Motor: Heat-Driven Linear Actuator - Hacktron Says It Used a libheif Bug and SSO Flaw to Reach OpenAI Internal Repos - Why Terence Tao Rejected the Fields Medallists’ AI Warning Letter - Unredacted filing alleges Microsoft and OpenAI treated AI scraping as theft - Qwen Launches Qwen3.8-Omni-Flash for Agentic Multimodal Tasks Episode Transcript OpenAI exploit chain raises alarms Let's start with security. Researchers at Hacktron say they chained a heap-overflow bug in libheif with a single-sign-on misconfiguration to compromise several OpenAI employees' ChatGPT and Codex accounts. According to their write-up, the entry point was OpenAI's community forum, where image uploads could trigger vulnerable HEIF processing. They say they limited their proof to a harmless internal pull request rather than touching sensitive code, and both OpenAI and Discourse moved quickly to patch the issues. Why this matters is bigger than one company: a narrow media-parsing bug, paired with weak identity boundaries, can become a route into high-value internal systems. It is also another sign that AI tools are making complex exploit chains quicker and cheaper to assemble. NYT lawsuit reveals AI concerns Staying with AI, the copyright battle between The New York Times, OpenAI, and Microsoft just got more serious. Newly unredacted court filings allege that senior executives privately described training on copyrighted news as theft and acknowledged that AI products could damage publishers by replacing visits to original articles. The filings also claim internal Microsoft data showed a major drop in click-throughs to The New York Times from Copilot. If those claims hold up, they could undercut the industry's public fair-use arguments and strengthen the case that the business impact on journalism was understood internally all along. This is becoming one of the most important legal fights in AI because it goes straight to the question of whether model training is innovation, appropriation, or something the courts will force into licensing. Tao reframes AI math risks In an update to the story we've been following about AI and mathematics, Terence Tao has explained why he did not sign a letter warning that AI could swamp the field with fast, low-effort results. Tao agrees there is a real problem, but he argues the issue is more nuanced. His view is that the bigger risk may not be an unreadable flood of theorems, but mathematicians gradually losing the hard problem-solving work that builds intuition in the first place. That is a subtle but important shift. It moves the conversation away from pure volume and toward what kinds of thinking researchers might stop practicing if AI becomes extremely capable. In other words, the concern is not just what AI produces, but what humans may stop doing. Hist