On mvanemmerik.com, Nov 16: NFL Week 10 chalk talk: Detroit dazzles, Cincinnati takes the North
vanemmerik.aiClaude, ChatGPT and Grok news, twice a day

Monday, November 16, 2026

ChatGPT for Healthcare adds agents that work across medical software

OpenAI’s healthcare update brings computer use and more than 20 plugins to care teams, with access and review controls still essential. It also announced research grants, outside safety assessments and a free abuse-image classifier after launching its science blog. Anthropic expanded older-model watermark coverage, while included Fable access for Pro and Team standard seats ends tonight and Cursor detailed its extension incident. No new xAI story cleared verification this evening.

Monday, November 16, 2026

OpenAIFeature

ChatGPT for Healthcare adds computer-use agents and more than 20 healthcare plugins

OpenAI says healthcare teams and eligible U.S. clinicians can now use computer-use agents through ChatGPT Work and Codex, with the capability also available in its API. ChatGPT for Healthcare gains more than 20 plugins and improved clinical search; access depends on workspace settings, and patient-data workflows require eligible services, a Business Associate Agreement and the required security settings.

Why it matters: Agents can move work between medical systems, but organizations still need to define access and human-review controls.

ChatGPTCodexOpenAI APIHappened Nov 16, 2026 · Sources: OpenAI healthcare announcement · OpenAI computer-use research and deployments

OpenAIResearch

OpenAI opens the next round of 50,000 academic research grants on November 23

OpenAI will award 50,000 grants to eligible researchers in application order, each providing free ChatGPT Plus access for a recipient and up to two collaborators. Applications open November 23; new grants use Plus instead of the pilot’s Pro access, and the program’s overall target rises to 200,000 researchers worldwide.

Why it matters: The expansion reaches more research teams while giving new recipients a lower subscription tier than the pilot.

ChatGPTOpenAIHappened Nov 16, 2026 · Sources: OpenAI researcher access announcement

OpenAISafety

OpenAI signs embedded safety-assessment agreements with Apollo, Goodfire, METR and Transluce

OpenAI says four outside organizations will assess alignment and safeguards with scoped access to internal models, data and systems. The agreements cover different evaluations rather than a completed safety certification; assessors may publish independently, while OpenAI retains confidentiality redactions and can request security-related disclosure delays.

Why it matters: The value of this deeper access will depend on the findings, their scope and what ultimately becomes public.

OpenAIHappened Nov 16, 2026 · Sources: OpenAI embedded assessment protocols

ClaudeSafety

Update: Anthropic lists text watermarks for more older Claude models, with sampling limits

Anthropic’s November 16 platform notes list text watermarking for Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 on supported platforms. Its support table warns that Amazon Bedrock rollout may take extra days and that non-default sampling settings can prevent reliable marking on these models; watermark detection remains a private preview.

Why it matters: Developers using older models cannot assume every generated passage will carry a detectable mark.

Try it: If you rely on the watermark, omit temperature, top_p and top_k, including values inserted by your client library.

Claude APIClaudeHappened Nov 16, 2026 · Sources: Claude Platform release notes · Claude marking coverage and limitations · Claude sampling guidance

OpenAISafety

OpenAI offers a free abuse-image classifier to vetted organizations

OpenAI has released an API classifier that flags potential child sexual abuse material, with a separate signal for material that may be AI-generated. Access is free for vetted organizations; the scores help prioritize human review rather than establish that an image depicts abuse.

Why it matters: Safety teams gain a tool for previously unseen images, but still need review and reporting processes.

OpenAI APIOpenAIHappened Nov 16, 2026 · Sources: OpenAI child-safety announcement · OpenAI API changelog

ClaudeFeature

Update: Included Fable access for Pro and Team standard seats ends tonight at 11:59 p.m. Pacific

Anthropic's launch offer lets Pro subscribers and Team standard-seat users spend up to half their weekly allowance on Fable models through November 16 at 11:59 p.m. Pacific. Afterward, Fable returns to separately billed usage credits on those plans; the offer's unused Fable reset also expires at the deadline.

Why it matters: The temporary access window is ending, so users should check which usage their subscription covers before continuing with Fable.

Try it: Check Settings > Usage for your remaining Fable allowance and any eligible reset; the deadline is 1:59 a.m. Central on November 17.

ClaudeClaude CodeHappened Nov 16, 2026 · Sources: Claude Help Center: Fable 5.2 launch offers

OpenAIResearch

OpenAI starts a science blog for research results, methods and independent researchers

OpenAI announced a science blog on November 15 for researchers to explain discoveries and how they use AI. It plans to publish scientific results, practical methods and researchers' perspectives, including ways to check outputs and the limits of those methods.

Why it matters: A methods-focused publication could help researchers assess and reproduce AI-assisted work; the announcement itself is not evidence that a scientific result is correct.

OpenAIChatGPTCodexHappened Nov 15, 2026 · Sources: OpenAI: Our new blog for science

ClaudeSafety

Update: Cursor traces the tampered Claude Code extension to an exposed vendor keyReported

In its November 13 investigation update, Cursor says a key exposed at Fireworks allowed access to coding traces containing credentials, leading to further intrusion. The attacker then put unauthorized Claude Code builds in Cursor's marketplace cache; Cursor says neither Anthropic's product nor Open VSX was compromised.

Why it matters: The new account identifies a vendor-to-marketplace attack path, rather than a flaw in Claude Code itself.

Try it: Affected users should follow Cursor's security notice and credential-rotation guidance, rather than assume that updating the extension alone removes the consequences.

Claude CodeHappened Nov 13, 2026 · Sources: Cursor: Incident investigation update · BleepingComputer: November 14 reporting