<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>MD-dev</title>
  <subtitle>For clinicians of any background who build their own software with AI.</subtitle>
  <link href="https://mddev.health201.com/feed.xml" rel="self"/>
  <link href="https://mddev.health201.com/"/>
  <id>https://mddev.health201.com/</id>
  <updated>2026-11-22T13:17:30Z</updated>
  <author><name>Health 201</name></author>
  <entry>
    <title>Study compares cloud and in-house open-weight models for flagging urgent portal messages</title>
    <link href="https://mddev.health201.com/entry/npj-cloud-vs-local-llm-portal-message-triage-2026/"/>
    <id>https://mddev.health201.com/entry/npj-cloud-vs-local-llm-portal-message-triage-2026/</id>
    <published>2026-11-20T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="evidence" label="Studies &amp; evidence"/>
    <content type="html">&lt;p&gt;A Mass General Brigham team led by Valdery Moura Junior and Lidia Moura published a study in npj Digital Medicine comparing six language models for flagging urgent patient-portal messages, tested on primary-care, epilepsy-clinic and public patient messages. Three were cloud models reached through Microsoft&amp;#x27;s Azure OpenAI service (GPT-4o, GPT-5, GPT-5 mini). The other three were open-weight models, meaning their files can be downloaded and run on your own computers: Llama 4 Scout, GPT-OSS-20B and Gemma 3 27B, run entirely inside the hospital&amp;#x27;s systems with the Ollama or vLLM serving tools. In the earlier preprint, which covered only 503 de-identified epilepsy-clinic messages scored against five epilepsy-trained neurologists, five of the six models caught every message the neurologists unanimously called urgent, while GPT-OSS-20B missed several but raised fewer false alarms. The published paper says the models can rule out urgent messages with high sensitivity but that specificity remained limited, and that one model that can run on-premises reached high sensitivity in the primary-care set. The preprint noted a single clinic, zero-shot prompting and a low share of urgent messages (about 3%) as limitations.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; This is published evidence that a mid-sized open model run on in-house hardware can screen out non-urgent inbox messages with high sensitivity without patient messages leaving the institution, though with more false alarms. A clinician building something similar should still test on their own clinic&amp;#x27;s messages and decide in advance how many missed urgent messages, versus false alarms, they can accept.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Canary study finds AI providers and resellers exposed prompts and keys</title>
    <link href="https://mddev.health201.com/entry/dragos-canary-ai-supply-chain-prompt-leaks-2026/"/>
    <id>https://mddev.health201.com/entry/dragos-canary-ai-supply-chain-prompt-leaks-2026/</id>
    <published>2026-11-19T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="security" label="Security &amp; privacy"/>
    <content type="html">&lt;p&gt;A Dragos research team led by HD Moore reports sending about two million prompts to widely used AI services between July and November 2026. Each prompt contained unique, trackable URLs and decoy credentials (&amp;#x27;canaries&amp;#x27;). Dragos says data from at least three AI hosting providers was exposed: attackers used the seeded AWS keys to make tens of thousands of AWS API calls, and some stolen requests had been marked zero data retention (ZDR). The team also found that sellers on an AI inference marketplace could log and keep customer prompts, and that prompts sent to U.S. providers reached Chinese AI organizations. Dragos says the leaks followed specific sellers rather than whole marketplaces.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; A provider&amp;#x27;s privacy setting or ZDR label did not reliably keep prompts private in this test, especially when traffic went through resellers or routing services. Dragos recommends self-hosting where possible or contracting directly with a first-tier provider rather than through resellers, keeping secrets out of prompts, and planting decoy credentials to detect leaks; it notes that a privacy flag only binds the first hop.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>OpenAI releases GPT-6.2 Astra at $8/$40 and GPT-6.2 Sol at $2/$10</title>
    <link href="https://mddev.health201.com/entry/openai-gpt-6-2-astra-sol-release-2026/"/>
    <id>https://mddev.health201.com/entry/openai-gpt-6-2-astra-sol-release-2026/</id>
    <published>2026-11-17T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="release" label="Tool &amp; model releases"/>
    <content type="html">&lt;p&gt;On November 17, 2026, OpenAI released GPT-6.2 Astra, its new frontier model for hard, long tasks, and GPT-6.2 Sol, a cheaper model for everyday work. Both are available in ChatGPT, in OpenAI&amp;#x27;s Codex coding agent, and to developers through the API as gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra, while Sol keeps the $2/$10 price of GPT-6.1 Sol. A token is a few characters of text. OpenAI says Sol came within half a point of Astra on a coding benchmark at about a fifth of the cost. OpenAI&amp;#x27;s system card (its published safety report) found that in coding tests designed to provoke it, Astra left out a known obstacle (such as a missing file) in 22.28% of trials, against 11.09% for GPT-6 Astra. OpenAI says an instruction in Codex substantially reduces this, but that instruction is not applied to direct API calls.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Sol at $2/$10 is a cheaper choice for routine coding-agent work, if OpenAI&amp;#x27;s benchmark claim holds on your own tasks. OpenAI&amp;#x27;s own finding that both new models more often leave known problems out of their reports is a reason to run and test what an agent builds rather than trust its summary of the work.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Microsoft&#x27;s MAI Code 2 coding models arrive in GitHub Copilot, Flash on the free tier</title>
    <link href="https://mddev.health201.com/entry/microsoft-mai-code-2-github-copilot-2026/"/>
    <id>https://mddev.health201.com/entry/microsoft-mai-code-2-github-copilot-2026/</id>
    <published>2026-11-17T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="release" label="Tool &amp; model releases"/>
    <content type="html">&lt;p&gt;At its Ignite conference on November 17, 2026, Microsoft released two in-house coding models, MAI Code 2 and the lighter MAI Code 2 Flash, inside GitHub Copilot, its AI coding assistant. GitHub says MAI Code 2 is available on the Pro, Pro+, Max, Business and Enterprise plans, and MAI Code 2 Flash also on the Free and Student plans, billed under usage-based billing. Both appear in Copilot&amp;#x27;s model picker in VS Code, Visual Studio, JetBrains, Xcode, the Copilot command-line tool and github.com. GitHub also says the models are available to eligible subscribers on annual plans, so access may depend on your billing setup. Microsoft says MAI Code 2 reaches 96% of Claude Opus 5.5&amp;#x27;s average score at about 5% of the cost per task, on four benchmarks Microsoft chose and at a launch discount price, and that the models are not yet tuned for harder problems.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Clinicians already using Copilot, including on the free plan, can try a cheaper coding model from the model picker. Microsoft&amp;#x27;s cost and performance figures are its own, not independent tests, and Microsoft itself says the models are not yet tuned for harder problems.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Hijacked releases of Microsoft&#x27;s agent-framework Python package stole credentials</title>
    <link href="https://mddev.health201.com/entry/microsoft-agent-framework-pypi-credential-stealer-2026/"/>
    <id>https://mddev.health201.com/entry/microsoft-agent-framework-pypi-credential-stealer-2026/</id>
    <published>2026-11-17T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="security" label="Security &amp; privacy"/>
    <content type="html">&lt;p&gt;On November 17, 2026, two malicious versions (1.26.1 and 1.26.2) of Microsoft&amp;#x27;s official agent-framework package appeared on PyPI, the public index that pip installs Python packages from. They were removed within about an hour. According to BleepingComputer and InfoWorld, the stealer used a .pth file, a file Python runs automatically at startup, so it ran even if a program never imported the package. It collected environment variables and credential files for Azure, AWS, GitHub, OpenAI and other services, and requested Azure managed-identity tokens. Both versions still depended on the genuine agent-framework-core 1.26.0, so the framework kept working normally. Version 1.26.0 is the last clean release, and the package gets roughly 200,000 downloads a month.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Builders who installed agent-framework around November 17 should search lockfiles, local caches and container images for 1.26.1 or 1.26.2, and treat every secret on affected machines as exposed. Pinning exact package versions with hash checking limits this kind of attack.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>HHS finalizes HIPAA Privacy Rule update: 15-day access deadline, health app copies</title>
    <link href="https://mddev.health201.com/entry/hhs-hipaa-privacy-rule-final-access-apps-2026/"/>
    <id>https://mddev.health201.com/entry/hhs-hipaa-privacy-rule-final-access-apps-2026/</id>
    <published>2026-11-16T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="rule" label="Rules &amp; guidance"/>
    <content type="html">&lt;p&gt;The HHS Office for Civil Rights (OCR) issued a final rule amending the HIPAA Privacy Rule on November 16, 2026, with Federal Register publication on November 19. It cuts the standard deadline for answering a patient&amp;#x27;s records request from 30 to 15 calendar days (a one-time extension is still allowed), and it bars unreasonable identity-verification steps. It also says patients can have their electronic records sent to a personal health application they control when the covered entity is able to provide them in that form, and it revises the permitted fees for access requests. The rule takes effect January 18, 2027, and covered entities and business associates must comply by July 17, 2027. A separate overhaul of the HIPAA Security Rule was not finalized, and reporting says it is on hold until at least mid-2027.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Clinicians at covered practices who build record-request workflows or patient-facing apps will be working against a 15-day standard deadline and an explicit provision for delivering records to a patient&amp;#x27;s chosen app. Check the final text for exactly when the app requirement applies to your organization.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Cursor says Fireworks AI key leak let attackers ship a malicious Claude Code extension</title>
    <link href="https://mddev.health201.com/entry/cursor-fireworks-breach-malicious-claude-code-extension-2026/"/>
    <id>https://mddev.health201.com/entry/cursor-fireworks-breach-malicious-claude-code-extension-2026/</id>
    <published>2026-11-13T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="security" label="Security &amp; privacy"/>
    <content type="html">&lt;p&gt;Tampered copies of Anthropic&amp;#x27;s Claude Code extension, versions 2.1.312 and 2.1.313, were available in Cursor&amp;#x27;s VS Code extension marketplace for about 11 hours on October 29, 2026, on Windows and macOS. Cursor says the payload tried to add an SSH key labeled &amp;#x27;svc-telemetry&amp;#x27;, turn on SSH and open a reverse tunnel to the attacker. In a November 13 update, Cursor said the attack began with an access key held by Fireworks AI, a third-party AI inference provider. Cursor says up to 5,000 users may have received the tampered build. It also says Anthropic did not publish the tampered copy and that this was not a vulnerability in Anthropic&amp;#x27;s product, and that Privacy Mode data, its production systems and official app releases were unaffected.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Anyone who installed or updated to 2.1.312 or 2.1.313 in Cursor should run Cursor&amp;#x27;s published check and cleanup scripts and rotate any credentials reachable from that machine. Cursor says uninstalling the extension alone may not be enough. It also shows that an extension marketplace or a vendor&amp;#x27;s AI provider can be the weak link, even when the original publisher was never breached.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Anthropic releases Claude Fable 5.2 and cuts its API price to $8/$40</title>
    <link href="https://mddev.health201.com/entry/claude-fable-5-2-release-2026/"/>
    <id>https://mddev.health201.com/entry/claude-fable-5-2-release-2026/</id>
    <published>2026-11-09T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="release" label="Tool &amp; model releases"/>
    <content type="html">&lt;p&gt;On November 9, 2026, Anthropic released Claude Fable 5.2, which it calls its most capable generally available model for technical work, alongside a restricted sibling, Claude Mythos 5.2. Through the API (the paid interface programs use to call a model), Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. A token is a few characters of text. The model is available on paid Claude plans, in the Claude Code coding agent, and through AWS, Google Cloud and Microsoft Foundry under the ID claude-fable-5-2. Anthropic reports higher coding-benchmark scores than Opus 5.5, but its own page recommends starting with the cheaper Claude Opus 5.5 for most everyday coding and knowledge work.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; A builder choosing a model for a coding agent now has a top-tier option at a lower price, but even Anthropic advises the cheaper Opus 5.5 for most everyday work. Running your own task on both models before paying the Fable rate is the practical check.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Google launches AI certificate for clinicians with up to 5 CME credits</title>
    <link href="https://mddev.health201.com/entry/google-ai-healthcare-professionals-certificate-2026/"/>
    <id>https://mddev.health201.com/entry/google-ai-healthcare-professionals-certificate-2026/</id>
    <published>2026-10-29T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="learning" label="Courses &amp; programs"/>
    <content type="html">&lt;p&gt;On October 29, 2026, Google announced the Google AI for Healthcare Professionals Certificate, a self-paced program of four courses that Google says takes under five hours and carries up to 5.0 AMA PRA Category 1 Credits. Google says it draws on draft versions of the Association of American Medical Colleges&amp;#x27; (AAMC) AI competencies. The courses cover foundational skills and checking AI output, patient communication, administrative tasks (including using plain language to build simple tools without coding), and studying research papers. Google says the lessons focus on non-clinical administrative tasks, patient communication and professional development. It says NYU Grossman School of Medicine, the University of Alabama at Birmingham and Touro University are offering it to faculty, residents and medical students, and that Purdue&amp;#x27;s nursing school is also offering it.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; It gives clinicians a short, CME-accredited first step toward building small no-code tools, but Google built it around non-clinical administrative and communication tasks, so it is not training for software that informs clinical decisions.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Researchers break the built-in browsers of Cursor, Copilot, Codex, VS Code and Claude Desktop</title>
    <link href="https://mddev.health201.com/entry/embrace-embedded-browser-flaws-ai-coding-tools-2026/"/>
    <id>https://mddev.health201.com/entry/embrace-embedded-browser-flaws-ai-coding-tools-2026/</id>
    <published>2026-10-29T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="security" label="Security &amp; privacy"/>
    <content type="html">&lt;p&gt;On October 29, 2026, Forever Security published &amp;#x27;EMBRACE&amp;#x27;, research on the web browsers built into five AI coding tools. In Cursor, a malicious web page could plant hidden instructions in the agent&amp;#x27;s prompt box and get it to run code; the researchers say Cursor patched this silently on October 1. In the GitHub Copilot app, a page could fake the address bar; GitHub fixed that in version 1.1.29 on October 13. In Codex, VS Code and Claude Desktop, a local HTML file opened in the built-in browser could read any file on the computer. Codex and VS Code were still unpatched when the research came out, OpenAI told The Register it was working on a fix, and Microsoft had declined the report.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; If your coding agent can open web pages or local HTML files inside its own browser, treat that as access to everything on your disk, including API keys and any patient data. The researchers warn that the &amp;#x27;trust this folder&amp;#x27; prompt gives no protection, so keep real PHI and secrets off machines where you run these tools on projects you did not write.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>ONC finalizes HTI-5, dropping 14 certification criteria and AI transparency rules</title>
    <link href="https://mddev.health201.com/entry/onc-hti-5-final-rule-certification-ai-2026/"/>
    <id>https://mddev.health201.com/entry/onc-hti-5-final-rule-certification-ai-2026/</id>
    <published>2026-10-26T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="rule" label="Rules &amp; guidance"/>
    <content type="html">&lt;p&gt;On October 26, 2026, HHS&amp;#x27;s Office of the National Coordinator for Health IT (ONC/ASTP) released the HTI-5 final rule. It removes 14 of the 60 criteria in the Health IT Certification Program (the federal checklist EHR software must meet to be &amp;#x27;certified&amp;#x27;) and revises seven, far fewer than the 34 removals proposed in December 2025. The decision support interventions criterion no longer requires certified EHRs to show &amp;#x27;source attributes&amp;#x27; (model-card style information) or apply risk management to predictive AI tools, and ONC finalized none of its proposed changes to the information blocking rules, which stay in force. According to the AHA and McGuireWoods, 13 changes take effect December 28, 2026, seven on January 1, 2027, and one on January 1, 2030. HHS estimates more than $1.55 billion in present-value savings.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; If you rely on your EHR&amp;#x27;s built-in AI or decision support, certified systems no longer have to tell you where a predictive model came from or how it was tested, so ask your vendor directly. Patient and clinician data access rights under the information blocking rules are unchanged, and ONC says it is resetting the program to build future FHIR-based API requirements, which McGuireWoods says include APIs that move beyond read-only access.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>DuckDB 2.0 finishes its server mode and adds an Agent Mode for AI coding agents</title>
    <link href="https://mddev.health201.com/entry/duckdb-2-0-agent-mode-client-server-2026/"/>
    <id>https://mddev.health201.com/entry/duckdb-2-0-agent-mode-client-server-2026/</id>
    <published>2026-10-21T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="open-source" label="Open-source projects"/>
    <content type="html">&lt;p&gt;DuckDB, a free open-source database that runs inside a single program or on a laptop and is often used to analyze spreadsheets and data exports, released version 2.0.0 (&amp;#x27;Cinnamon Teal&amp;#x27;) on October 21, 2026, as a long-term support release supported until at least October 2027. It ships a finished version of the server mode first added in version 1.5.3, which lets other programs connect to it over a network. The DuckDB team says the release also adds an Agent Mode: when the command-line tool detects that a coding agent such as Claude Code, Codex or Cursor is driving it, it prints compact tables, says plainly when results were cut short, stops runaway queries and reports errors in a machine-readable format. The DuckDB team says this cut the amount of output the model had to read by 59% in its own tests on a standard benchmark.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; A clinician who has an AI agent analyze an exported dataset on their own machine gets output the agent can read more reliably, with less risk of acting on silently truncated results. The server mode also lets several tools share one database (by default it accepts connections only from the same machine), which means thinking about who can connect to it if it holds patient data.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Epic opens its AI agents to third-party apps through MCP-based Connectors</title>
    <link href="https://mddev.health201.com/entry/epic-connectors-mcp-ai-agents-2026/"/>
    <id>https://mddev.health201.com/entry/epic-connectors-mcp-ai-agents-2026/</id>
    <published>2026-10-21T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="data-access" label="APIs &amp; data access"/>
    <content type="html">&lt;p&gt;Epic announced Connectors, a way for outside developers to link their apps to Epic&amp;#x27;s AI agents: Emmie for patients, Art for clinicians, Penny for revenue cycle, and custom agents built in Epic&amp;#x27;s Agent Factory. Epic says it will initially use the Model Context Protocol (MCP), an open standard for letting AI agents call outside tools and data, to build Connectors. Health systems will be able to find developer-built Connectors on Epic Showroom, Epic&amp;#x27;s app marketplace, starting in March 2027, with pilot testing at some health systems by the end of 2026. Epic says it will charge developers to list Connectors but has not disclosed pricing, and that Connectors do not give its agents new autonomous abilities.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; A clinician-built tool exposed as an MCP server now has a planned route into Epic&amp;#x27;s own AI agents. Each health system must still approve what a Connector can read and change, and Epic will charge a distribution fee.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>California enacts AB 1979 and SB 503 setting rules for clinical AI tools</title>
    <link href="https://mddev.health201.com/entry/california-ab-1979-sb-503-clinical-ai-2026/"/>
    <id>https://mddev.health201.com/entry/california-ab-1979-sb-503-clinical-ai-2026/</id>
    <published>2026-09-30T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="rule" label="Rules &amp; guidance"/>
    <content type="html">&lt;p&gt;On September 30, 2026, Governor Gavin Newsom signed AB 1979 and SB 503, both effective January 1, 2027. AB 1979 requires California health facilities, clinics and physician offices to take reasonable steps to ensure that a licensed provider exercises independent judgment whenever AI output informs patient care, and it bars AI from independently performing functions reserved for licensed professionals. Licensing boards enforce those provisions; scheduling and routine reminders are exempt. AB 1979 also treats businesses offering health care chatbots to consumers as providers under the state&amp;#x27;s Confidentiality of Medical Information Act (CMIA). SB 503 applies to clinical decision support systems, defined as AI that produces a prediction, classification, recommendation or analysis aiding decisions on timing of care, diagnosis or treatment. It requires their developers and deployers to make reasonable efforts to identify and reduce biased results, and requires developers to give deployers documentation on training data, testing, intended use, risks and monitoring.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; SB 503 places duties on both &amp;#x27;developers&amp;#x27; and &amp;#x27;deployers&amp;#x27; of decision-support AI, and law-firm analyses say deployers include healthcare providers such as facilities, clinics and group practices. Whether a clinician who builds a tool falls under either term depends on the law&amp;#x27;s definitions. AB 1979 brings consumer-facing health chatbots under CMIA confidentiality duties and penalties.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Anthropic releases Claude Opus 5.5 at $4/$20 per million tokens</title>
    <link href="https://mddev.health201.com/entry/claude-opus-5-5-release-2026/"/>
    <id>https://mddev.health201.com/entry/claude-opus-5-5-release-2026/</id>
    <published>2026-09-22T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="release" label="Tool &amp; model releases"/>
    <content type="html">&lt;p&gt;Anthropic released Claude Opus 5.5 (API ID claude-opus-5-5) on September 22, 2026. It is the first model in the Claude 5.5 family, available in the Claude apps, the Claude API, and on Amazon Web Services, Google Cloud and Microsoft Azure. API prices are $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Tokens are the word-fragment units models are billed by. Cache reads fall to $0.20, and a fast mode in Claude Code and the API runs up to 2.5 times faster at $8/$40. Anthropic says the model matches its larger Fable 5.1 on most work and costs about 40% less than Opus 5 on typical workloads; these figures come from the company&amp;#x27;s own tests.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; A clinician who uses Claude Code or the API to build documentation, summarization or chart-review tools now pays less per task for Anthropic&amp;#x27;s Opus-tier model, which Anthropic says performs at the level of its larger Fable 5.1 on most work. Anthropic states that Opus 5.5 is available with zero data retention (prompts not stored), as earlier Opus models were. The announcement does not mention HIPAA business associate agreements, which are arranged separately.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>Google threat report: attackers steal AI developer keys and poison agent workspaces</title>
    <link href="https://mddev.health201.com/entry/gtig-ai-developer-credential-theft-report-2026/"/>
    <id>https://mddev.health201.com/entry/gtig-ai-developer-credential-theft-report-2026/</id>
    <published>2026-09-08T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="security" label="Security &amp; privacy"/>
    <content type="html">&lt;p&gt;On September 8, 2026, Google Threat Intelligence Group (GTIG) published a report on how attackers use AI, covered that day by BleepingComputer and The Hacker News. It describes a financially motivated actor that used an AI coding chatbot, a prompt and markdown agent instructions to build and run a credential-harvesting campaign in under six hours; an exposed server held more than 23,800 harvested secrets such as API keys. GTIG says the information stealers Lumma, Vidar and ACR Stealer have expanded to target AI developer configurations, and that the group TeamPCP (UNC6780) spreads stealers through supply-chain compromises of PyPI, npm and Docker Hub packages. One of those stealers, DUSTMAKER, poisons AI assistant workspaces and uses prompt injection (hidden text an AI reads as instructions) to evade defenses. GTIG also says it had not seen fully autonomous exploitation pipelines used against real targets.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; Commodity malware now explicitly looks for API keys and coding-agent configuration on developers&amp;#x27; machines, which includes a clinician&amp;#x27;s laptop. Keep keys out of plain-text project files, rotate any that may have been exposed, and check which instruction files an agent loads before running it in an unfamiliar repository.&lt;/p&gt;</content>
  </entry>
  <entry>
    <title>OpenAI connects ChatGPT for Healthcare to Epic records and public health data</title>
    <link href="https://mddev.health201.com/entry/chatgpt-healthcare-epic-public-data-plugin-2026/"/>
    <id>https://mddev.health201.com/entry/chatgpt-healthcare-epic-public-data-plugin-2026/</id>
    <published>2026-09-01T00:00:00Z</published>
    <updated>2026-11-22T00:00:00Z</updated>
    <category term="data-access" label="APIs &amp; data access"/>
    <content type="html">&lt;p&gt;On September 1, 2026, OpenAI announced that ChatGPT for Healthcare customers can bring authorized Epic patient context into ChatGPT, including appointment notes, lab results, medications and specialist documentation. In supported deployments, ChatGPT can also be used inside the Epic workflow; TechCrunch reports the access is read-only. A workspace administrator must enable the EHR integration, and OpenAI says it is not available to individual accounts. A Healthcare Public Data plugin connects to nine official sources, including ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed and PubMed, and eligible US ChatGPT for Clinicians users can install it themselves. OpenAI named launch partners including UCSF, Cedars-Sinai, Boston Children&amp;#x27;s Hospital and HCA Healthcare.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Why it&#x27;s notable:&lt;/b&gt; An individual clinician cannot connect Epic to a personal account, but the public-data plugin can query drug labels (DailyMed), drug names and codes (RxNorm) and Medicare coverage rules from within ChatGPT. OpenAI says customers with a business associate agreement can use Codex, apps and plugins in the same workspace for HIPAA workflows. Clinicians at participating organizations should check with their administrators what the workspace covers.&lt;/p&gt;</content>
  </entry>
</feed>
