1. Scott Shambaugh Closed a Bot's Pull Request. The Bot Wrote an Article Attacking Him. Scott Shambaugh reviews code for matplotlib, the Python plotting library downloaded roughly 130 million times a month.

2. OpenAI Erased "Safely" from Its Mission While xAI Pushed Grok to Go Wilder OpenAI's 2023 IRS tax filing described its mission as building AI "that safely benefits humanity, unconstrained by a need to generate financial return.

3. AI Splits Software Engineering Into Two Tracks Spotify's top developers haven't written a line of code since December.

In Brief

• OpenAI Claims GPT-5.2 Derived a New Result in Theoretical Physics OpenAI published a preprint with researchers from the Institute for Advanced Study, Vanderbilt, Cambridge, and Harvard showing that certain gluon amplitudes previously assumed to be zero can arise under specific conditions. An internal scaffolded version of GPT-5.2 spent roughly 12 hours reasoning through the problem and produced a formal proof matching the original conjecture. The team has already extended the result from gluons to gravitons.

• Google Upgrades Gemini 3 Deep Think for Science and Engineering Google released a major update to Gemini 3 Deep Think, its specialized reasoning mode, which solved 18 previously unsolved research problems and disproved a decade-old mathematical conjecture. The model scored 48.4% on Humanity's Last Exam, 84.6% on ARC-AGI-2, and reached Legendary Grandmaster on Codeforces. Deep Think is now available via the Gemini API to select researchers and enterprises for the first time.

• OpenAI Ships GPT-5.3-Codex-Spark on Cerebras Hardware at 1,000 Tokens per Second OpenAI released Codex-Spark, a smaller sibling of GPT-5.3-Codex optimized for real-time coding on Cerebras' Wafer Scale Engine 3. Spark runs 15x faster than Codex 5.3 but scores 16 points lower on SWE-Bench Pro. The model is available as a research preview to ChatGPT Pro users.

• Hollywood Studios Send Cease-and-Desist Letters to ByteDance Over Seedance 2.0 ByteDance's new AI video generator Seedance 2.0 drew immediate backlash from Hollywood after users created viral clips featuring Spider-Man, Darth Vader, and real actors without authorization. Disney and Paramount sent cease-and-desist letters, and SAG-AFTRA condemned the tool alongside the studios. ByteDance said it will strengthen safeguards to prevent unauthorized use of intellectual property and likenesses.

• OpenAI Adds Lockdown Mode to ChatGPT to Block Prompt Injection Exfiltration OpenAI introduced Lockdown Mode, an optional security setting that disables web browsing, image rendering, Deep Research, and Agent Mode to prevent prompt-injection-based data exfiltration. The feature targets high-risk users such as executives and security teams at prominent organizations. Lockdown Mode is available now on Enterprise, Edu, Healthcare, and Teachers plans, with consumer rollout planned later.

• 241 News Sites Block Internet Archive Crawlers Over AI Scraping Fears Major publishers including The New York Times, The Guardian, and the Financial Times have blocked Internet Archive's crawlers from accessing their content. Publishers call the Wayback Machine a "back door" that lets AI companies scrape their archives without authorization. The Guardian admits it has not documented any AI company actually scraping its content through the Archive.

• Anthropic Appoints Former Microsoft CFO Chris Liddell to Board Anthropic named Chris Liddell its sixth board member, joining CEO Dario Amodei, President Daniela Amodei, and Netflix Chairman Reed Hastings. Liddell previously served as CFO of Microsoft and General Motors and as Deputy White House Chief of Staff under Trump's first administration. He oversaw GM's $23 billion IPO in 2010, the largest public offering at the time.

• India Approves $1.1B State-Backed Fund-of-Funds for Deep Tech and Manufacturing India's government approved a $1.1 billion fund-of-funds that will invest through private venture capital firms to back deep-tech and manufacturing startups.

• Stoat Strips All LLM-Generated Code After Users Demand Transparency Open-source messaging project Stoat removed all LLM-generated code from its codebase after users discovered Claude listed as a contributor and demanded disclosure. The developers published a formal policy and publicly reverted the AI-written portions.

• OpenAI Open-Sources GABRIEL Toolkit for Social Science Research OpenAI's economics team released GABRIEL, an Apache 2.0-licensed Python toolkit that uses GPT to convert qualitative text and images into quantitative data. The library handles dataset merging, deduplication, passage coding, and de-identification, returning structured DataFrames. Available now via pip install openai-gabriel.

• Anthropic Partners with CodePath to Bring Claude to 30,000 CS Students Anthropic partnered with CodePath, the largest collegiate computer science program in the U.S., to integrate Claude into its curriculum.

Read the full edition →