New mystery Claude model appears in testing; OpenAI adds OpenClaw to ChatGPT
🤔 “How do I set up my OpenClaw agent?” Share Handy AI with your coworkers and friends to help them understand the crazy world of modern artificial intelligence (and save you some time).
what to know for now
🪐 Anthropic is red-teaming a model codenamed “claude-jupiter-v1-p” ahead of the May 6 conference. Testing Catalog reported the internal exercise running this week. Anthropic ran a similar pre-launch red-team under the codename “Neptune” before the Claude 4 family launched. Make of that what you will. Read more
🤖 OpenAI added ChatGPT subscription access to OpenClaw. The open-source AI agent framework (3.2 million users) now lets ChatGPT Plus subscribers sign in via OAuth and run autonomous coding agents through the Codex endpoint using GPT-5.4. Anthropic blocked Claude subscriptions from the same integration. Read more
🪖 Anthropic is the only major AI lab that wouldn’t sign the Pentagon’s classified network deal. The Department of War announced agreements with seven vendors (Microsoft, Google, OpenAI, AWS, Nvidia, xAI/SpaceX, and Oracle) to deploy AI on its Impact Level 6 and 7 classified networks. Hundreds of Google employees signed an open letter calling the contract “inhumane.” Their employer signed it anyway. Read more
⚖️ Musk admitted under oath that xAI trained Grok by distilling OpenAI’s model outputs. During testimony at the Musk v. Altman trial in Oakland, he acknowledged xAI used model distillation from OpenAI outputs during Grok’s development and framed it as a general industry practice. He is suing OpenAI for betraying its nonprofit mission. Per his own testimony, he was also cloning their model. Read more
🧌 ChatGPT developed an involuntary goblin fixation. After the GPT-5.1 launch, goblin references in responses spiked 175%. OpenAI traced it to the “Nerdy” personality setting, which generated 66.7% of all goblin mentions despite representing just 2.5% of responses; a reward signal reinforced creature-word outputs and spread the behavior model-wide. Read more
🔒 Anthropic launched Claude Security in public beta. Powered by Claude Opus 4.7, it scans entire codebases, traces data flows, explains its exploitability assessments, and can apply patches directly inside a Claude Code session. Integration partners include CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, and Wiz; available to all Enterprise customers now. Read more
🩺 Mayo Clinic’s AI model detects pancreatic cancer on routine CT scans up to three years before clinical diagnosis. Published in Gut, the validation study for REDMOD flagged 73% of pre-diagnostic cancers at a median of 16 months before symptoms appeared, nearly doubling specialist radiologist detection rates on the same scans. Pancreatic cancer’s five-year survival rate sits around 13% when caught late. Read more
⚖️ Elon Musk’s $130 billion lawsuit against OpenAI went to trial in Oakland. Jury selection started April 27 before Judge Yvonne Gonzalez Rogers. Musk is demanding the unwinding of OpenAI’s 2025 conversion to a public benefit corporation, alleging it betrayed the original nonprofit mission. Week one included three days of Musk testimony; the judge expects a ruling by mid-May. Read more
🧪 AI Research of the Week
AI Outperforms Attending Physicians in ER Diagnosis Study
From Harvard Medical School and Beth Israel Deaconess Medical Center
Jake’s Take: Researchers at Harvard and BIDMC tested OpenAI’s o1-preview model on real ER triage cases and compared it against two attending physicians from elite medical institutions, using only text-based clinical inputs (no imaging, no physical exam, just the notes a triage nurse would hand off). The AI correctly identified the exact or near-correct diagnosis in 67% of cases, against 55% and 50% for the two physicians, a gap that held across common presentations and rare or complex cases alike.
It’s a great study by the numbers, but the really interesting bit here is that o1-preview is an 18-month-old model. Whatever gap existed then is wider now, and the frontier has not been narrowing in medicine’s favor. The researchers call for prospective trials before clinical deployment, which is the right call.
what to know for later
📜 OpenAI and Microsoft rewrote their partnership and removed the AGI clause. The original deal would have voided entirely if OpenAI declared it had achieved AGI, a clause that gave OpenAI enormous leverage over its biggest investor. The new agreement eliminates it, ends Azure exclusivity so OpenAI can sell through AWS and Google Cloud, and gives Microsoft a 20% revenue share through 2030. Read more