# Anthropic Life Sciences Verification Program

> Source: https://aiwiki.ai/wiki/anthropic_life_sciences_verification_program
> Updated: 2026-09-29
> Fact-checked: 2026-09-29
> Categories: AI Safety, Anthropic, Drug Discovery
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Anthropic Life Sciences Verification Program." aiwiki.ai, 29 Sept 2026. https://aiwiki.ai/wiki/anthropic_life_sciences_verification_program
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

The **Life Sciences Verification Program** (**LSVP**) is a trusted-access program run by [Anthropic](https://aiwiki.ai/wiki/anthropic) that lets vetted life sciences organizations use its Claude models with biology safeguards that are more permissive than those on its generally available models. Anthropic first named the program on September 1, 2026, alongside the launch of [Claude Fable 5.1](https://aiwiki.ai/wiki/claude_fable_5_1) and [Claude Mythos 5.1](https://aiwiki.ai/wiki/claude_mythos_5_1), as an invite-only beta built "in partnership with the US government."[4][5] On September 17, 2026 it opened applications to the wider life science community, with two kinds of access grant ("Standard Use" and "High-risk Use") covering Mythos, Opus and Sonnet models.[1] The program relaxes real-time blocking of individual biology requests and, in Anthropic's words, shifts safeguards toward offline monitoring of each organization's traffic against the use cases it declared; in exchange it requires 30-day data retention for program usage.[1][2] As of September 29, 2026 the program was still in beta and open only to organizations on Anthropic's first-party Enterprise, Team and API offerings.[2]

## Background

Anthropic's most capable Claude 5 models ship in two configurations of the same weights: a generally available "Fable" model with extra safeguards and a restricted "Mythos" model.[20] Anthropic describes Fable 5.1 and Mythos 5.1 as "the same model, but with different levels of safeguards"; Mythos 5.1 "is available only through our trusted access programs; its safeguards are specifically designed to support work in cybersecurity and the life sciences."[4] On the generally available side, safety classifiers watch for biology requests and, when one fires, hand the request to a less capable model. Anthropic's help center says Fable 5 and Fable 5.1 fall back on "a large fraction of queries we consider dual-use in biology, such as virology, toxicology, drug design, and molecular design," which is why it does not recommend them "for professional biology research and drug development at this time"; flagged biology, chemistry and life sciences requests currently fall back to [Claude Opus 5](https://aiwiki.ai/wiki/claude_opus_5).[19]

On August 7, 2026 Anthropic retrained the [Claude Fable 5](https://aiwiki.ai/wiki/claude_fable_5) biology classifier, saying the update cut biology-related fallbacks "by about 85%" across its products. The same post said Fable still fell back to Opus 5 for dual-use requests "including virology, toxicology, and molecular design," so "it isn't yet usable for professional biology research and drug development," and that Anthropic was "committed to closing that gap through trusted access pathways for frontier biology capabilities."[7] The company explained its caution by pointing to the dual-use nature of biology: the capabilities that help a researcher develop a treatment could, "in the wrong hands," help a malicious actor develop a biological weapon, and its assessments showed Fable 5 "could provide significant uplift to such an actor."[7]

## Timeline

| Date (2026) | Event |
| --- | --- |
| June 9 | At the Fable 5 and Mythos 5 launch, Anthropic says it plans to open "a trusted access program for biology" giving a small number of researchers Fable 5 "with the biology and chemistry safeguards removed (but the cyber safeguards still in place)."[29] |
| August 7 | Anthropic loosens Fable 5's biology classifier and commits to "trusted access pathways" for professional biology work.[7] |
| August 27 | In "Expanding our support for scientists," Anthropic says it is "working in partnership with the US government to establish an access program for life sciences professionals to use Mythos-class models" and has "now enrolled our first participants."[8] |
| September 1 | The Fable 5.1 and Mythos 5.1 launch names the Life Sciences Verification Program. Anthropic's Mythos page describes it as "launching as an invite-only beta."[4][5] |
| September 10 | Anthropic publishes its September 2026 threat intelligence report, including five case studies of biological misuse, which the later LSVP announcement cites.[9][25] |
| September 16 | Claude Science version 0.1.49 lets members of LSVP organizations choose a use case for each session.[12] |
| September 17 | Anthropic opens applications to the broader life science community and publishes the Standard Use and High-risk Use grant structure.[1][21] |
| September 22 | [Claude Opus 5.5](https://aiwiki.ai/wiki/claude_opus_5_5) launches with Fable 5.1-level biology safeguards; Anthropic invites vetted organizations to apply to the LSVP to use it for biology research.[13] |
| September 28 | [Claude Sonnet 5.5](https://aiwiki.ai/wiki/claude_sonnet_5_5) launches with the same biology safeguards as Sonnet 5; the help center says the LSVP relaxes them on Sonnet 5.5 only through the High-risk Use add-on.[15][17] |

## Early-access phase (to September 16)

The September 1 launch post said Anthropic had "established an access program, developed in partnership with the US government, to enable access to Claude Mythos 5.1's advanced biology capabilities" and expected "to open enrollment for scientists soon."[4] It described the LSVP as a program under which "life sciences professionals can use Claude Mythos 5.1 with safeguards designed for professional research and development activities (while all other safeguards remain in place)," and said: "In partnership with the US government, we have enrolled our first participants."[4] At the time Mythos 5.1 was "only available to a set of US organizations," with Anthropic "coordinating with the US government to expand access to a broader set of domestic and international partners."[4] The accompanying system card said Mythos 5.1 relaxes "life sciences safeguards through our Life Sciences Verification Program."[6]

Writing on launch day, R&D World observed that Anthropic had "yet to identify the participating agency, publish eligibility criteria or give a date for broader applications."[22] The September 17 announcement later described the vetting criteria and opened applications. It says Anthropic is "working with the US government" but does not name an agency.[1]

## Program design (September 17 onward)

Anthropic's September 17 announcement said it had "already onboarded dozens of organizations through an early-access program" and was "now opening applications to the broader life science community," launching the program "in beta, initially for teams and institutions."[1] It said the program is meant for tasks "currently blocked in our generally available Fable models, like drug discovery, research biology, clinical development, and manufacturing," and is built "for teams of all kinds," from academic labs to startups and pharmaceutical companies.[1] Anthropic's official X account announced that life scientists could use its models "including, for the first time, Mythos."[21]

### Verification

Before receiving any grant, each applicant goes through "a verification process that includes a review of their research credentials, security standards, and ethical research oversight."[1] The help-center article sets out the steps as of September 29, 2026:[2]

1. The organization submits an interest form (organizations already invited by their Anthropic account team skip this step). The form page says completing it "does not guarantee your organization access."[3]
2. Anthropic reviews the interest. "Not every organization that submits the form will be invited during the beta."
3. Anthropic opens an application for the organization in its Verification Portal. The person submitting must verify their identity.
4. An Admin, Owner or Primary Owner of the organization applies. The application "takes about 15 minutes, starting with a business verification step."
5. Anthropic reviews it; decisions "typically come within about three business days."
6. On approval, the grant is provisioned to the organization automatically.

The application asks for organization details, "one life sciences reference, such as an FDA registration, a clinical trial ID, or a public grant number," a description of the team and what it will use Claude for, the name of the IT admin, and attestations "including that use is internal to your organization." Anthropic says it "doesn't ask for program-level science, targets, or readouts."[2] Program requirements are 30-day data retention, "a dedicated workspace, with phishing-resistant multi-factor authentication for the Claude apps or Workload Identity Federation for the API," and use by the organization "for internal work only."[2] Each approved team (each team is its own grant) gets up to 100 seats, and an organization may apply for several grants.[2]

### Grant types

| | Standard Use | High-risk Use |
| --- | --- | --- |
| Purpose | "Most life science work, including the majority of biology research and development workflows"[1] | "An add-on grant for teams working in areas blocked under Standard Use"[1] |
| Scope | Can be extended to entire teams for daily work[1] | A single research project rather than a whole team[1] |
| Renewal | Once a year[1] | Every six months[1] |
| What changes | Access to Mythos, Opus and Sonnet models "with refined classifiers that are more permissive for science tasks than our generally available models"[1] | "Removes all safeguards that block life sciences requests"[1] |
| Availability at launch (Sept. 17) | Mythos 5.1, Opus 5 and Sonnet 5, and future models as they launch[1] | Opus 5 and Sonnet 5; for Mythos, limited "to a small set of entities with additional vetting"[1] |

Anthropic's example of a High-risk Use project is "characterizing how one specific family of viral vectors is recognized by human immune pathways." It expects a researcher doing dual-use work to hold one Standard Use grant for daily activities plus one or more project-specific High-risk grants.[1] The help center lists "virology or toxin research" as examples of work that may need the add-on.[2] Anthropic said it was "working with the US government to make high-risk grants more broadly available for Claude Mythos."[1] Under both grants, "all other safeguards, such as cyber classifiers, will remain in place."[1] Cybersecurity access is handled by a separate program, the [Cyber Verification Program](https://aiwiki.ai/wiki/anthropic_cyber_verification_program).[4]

The Standard Use grant is meant to cover work "spanning basic science, R&D, supply chain and manufacturing, clinical development, quality assurance, regulatory affairs, investing and diligence, and more."[1] The help center lists eleven supported use cases:[2]

- Drug and countermeasure discovery and preclinical support
- Understanding how diseases work
- Clinical development and operations
- Bioinformatics, computational biology, and research data infrastructure
- Diagnostics and laboratory medicine
- Manufacturing, quality, and supply chain for regulated products
- Regulatory affairs, medical affairs, and market access
- Public health, biosurveillance, biosafety, and biosecurity
- Agricultural, veterinary, and environmental biosciences
- Life sciences funding and diligence
- Life sciences education and training

Neither the announcement nor the help-center article lists a separate price for LSVP grants; MIXED noted that "no pricing is given."[1][2][24]

## Models covered

The set of models under the program has changed with each Anthropic release.

| Model | LSVP status | Source |
| --- | --- | --- |
| Claude Mythos 5.1 | Standard Use from launch; High-risk Use limited to a small set of additionally vetted entities | [1][2] |
| [Claude Opus 5](https://aiwiki.ai/wiki/claude_opus_5) | Standard Use and High-risk Use from September 17 | [1] |
| [Claude Sonnet 5](https://aiwiki.ai/wiki/claude_sonnet_5) | Standard Use and High-risk Use from September 17 | [1] |
| Claude Opus 5.5 | Anthropic invited vetted organizations to apply to use it for biology research at its September 22 launch; listed among covered models by September 29 | [2][13] |
| Claude Sonnet 5.5 | Biology safeguards relaxed "only through the High-risk Use add-on" | [2][17] |
| Claude Fable 5 and Fable 5.1 | Not part of the program; "keep their full biology safeguards under every grant" | [2] |
| Claude Mythos 5 (preview) | "Not offered under LSVP access" | [2] |

The help center tells grant holders to "use Claude Mythos 5.1 instead of Fable."[2] The interest-form page, as of September 29, described the program as giving "expanded access to Mythos 5.1 for biology and drug discovery, and Opus 5.5 and Sonnet 5 with a refined bio safety classifier."[3]

The Opus 5.5 launch post said that because Opus 5.5 is "comparable to Claude Mythos 5.1 in biology and cybersecurity," it uses the same biology safeguards as Fable 5.1, and that users whose research is "impeded by these safeguards" can apply to "our new Life Sciences Verification Program, which gives vetted organizations like academic labs, startups, and pharmaceutical companies access to safeguards designed for the full breadth of biology-related work."[13] Its system card says Anthropic treats Opus 5.5 as having CB-1 but not CB-2 chemical and biological capabilities.[14] Anthropic's help article on Opus fallbacks also tells organizations doing "legitimate life sciences research" that are affected by the safeguards to apply to the LSVP.[18] Sonnet 5.5 takes a different approach: its system card says Anthropic released it "with the same safeguards we have deployed for Opus 5, rather than the broader dual use research biology classifiers used for Opus 5.5."[16] The Sonnet 5.5 launch post says these safeguards leave "most research, education, and clinical work" unaffected, "though some microbiology and virology requests may be flagged in error," and points organizations to the LSVP.[15] On Sonnet 5.5, biology blocks do not fall back to another model; the request is blocked outright.[17]

## Monitoring and data retention

The LSVP's safeguards are built around what Anthropic calls "shared responsibility." Because it vets participating organizations "for their life sciences credibility and oversight," Anthropic says it can let them "specify for themselves what constitutes safe usage for teams or projects within their program."[1] Anthropic says the design was made "in close collaboration with enterprise CISOs" and is aimed at three threat models:[1]

| Threat model | Anthropic's description |
| --- | --- |
| Access compromise | "Malware or account takeover diverting access to a bad actor" |
| Insider threats | "Rogue or coerced employees intentionally taking malicious action or diverting their access to a bad actor" |
| Agent misuse | "Agents, especially working in swarms or over long-horizon tasks, taking unintended dangerous actions" |

Each organization's access is tied to the use cases in its grant application, which should be "high-level descriptions of the intended work, like one would share in a job listing," without sensitive information or intellectual property. Anthropic says it continuously monitors LSVP traffic for usage outside that scope and can flag cases to organization admins, who act "within pre-agreed timeframes."[1] The help center says that if something looks out of scope, Anthropic will "notify your admin directly and give you one week to explain, narrow the scope, or dispute it."[2]

For LSVP traffic, Anthropic says it is "shifting safeguards from real-time blocking, where we reject potentially harmful access at the time of each request, to offline monitoring," because "serious misuse is often spread across many requests and sessions to look disconnected and evade detection."[1] That shift is the stated reason for retention. Program usage is kept for 30 days, with "automated safety monitoring and human review of flagged activity in rare circumstances."[2] Zero data retention is not available under a grant, and existing zero-retention or custom-retention agreements do not apply to LSVP usage. Retention covers only the program workspace or organization.[2] Anthropic says the data "is strictly compartmentalized and cannot be used for model training or accessed by members of Anthropic's life sciences research teams," and the help center adds that it is not "analyzed for any commercial or research purpose."[1][2]

The retention rule follows a wider Anthropic policy. Since June 9, 2026, prompts and outputs of Mythos-class "covered models" have been retained for 30 days for safety work, including for customers who otherwise have zero data retention, although since September 2026 some eligible zero-retention customers have received ZDR on Fable 5 and Fable 5.1 until Enterprise Frontier Safeguards are ready.[20][10] Anthropic's Enterprise Frontier Safeguards (EFS), announced September 1, 2026, are meant to combine zero-retention privacy with misuse detection by storing data in infrastructure the customer controls.[10] The LSVP announcement says Anthropic is "working to understand how LSVP can integrate with features from our Enterprise Frontier Safeguards (EFS) systems" for organizations that qualify.[1]

## Availability and product surfaces

According to the announcement, grants work on all Anthropic product surfaces, "including Claude Science, Claude.ai, Claude Code and the API."[1] At launch the LSVP was available in Anthropic's first-party console for API usage and in Claude for Enterprise and Team plans, but not on individual plans or third-party platforms.[1] The help center, updated on September 29, 2026, lists these limits for the beta:[2]

- No Pro or Max plans. Anthropic says it plans to expand to individual Pro and Max plans over time, but the help center also says access is granted "to a verified organization, not to a person," and individuals not affiliated with an organization are not eligible.[1][2]
- No access for customers who use Claude only through a cloud provider (Amazon Web Services, Google Cloud or Microsoft Azure).
- No workloads involving protected health information. The program is not covered by a Business Associate Agreement under [HIPAA](https://aiwiki.ai/wiki/hipaa), so organizations set up under a BAA cannot receive grants; they can apply through a separate non-BAA organization for non-PHI work.[1][2]
- Platforms cannot pass access through to their own customers, though a platform can apply for its own internal R&D.

Switching between grants differs by product. In the API and Claude Science, users can switch grants natively. In Claude.ai and [Claude Code](https://aiwiki.ai/wiki/claude_code), only a preselected default grant applies at first, except when Claude Code runs with API authentication.[1]

[Claude Science](https://aiwiki.ai/wiki/claude_science), Anthropic's desktop research application, has its own LSVP controls.[11][27] If an organization has approved several use cases, users pick one per session, and Claude Science "sends a session's use case with the requests it makes for that session, so Claude's safeguards can apply the access approved for that work."[11] The documentation says that under the program "protections against the most serious misuse stay in place, and safeguards can still flag a message."[11] The feature arrived in version 0.1.49 on September 16, 2026, the day before public applications opened. Version 0.1.51 (September 22) fixed a case where approved use cases could disappear from the app.[12]

## Relationship to other Anthropic science efforts

The LSVP is an access and safeguards layer, not a product. It sits alongside Anthropic's other life sciences work:

- **[Claude for Life Sciences](https://aiwiki.ai/wiki/claude_for_life_sciences)** was announced on October 20, 2025 as a set of improvements meant to make Claude useful "from early discovery through to translation and commercialization."[28] LSVP grants change which biology requests the underlying models will answer for verified organizations.
- **Claude Science** is a desktop application that "pairs Claude with an analysis environment on your computer."[27] It is one of the surfaces where LSVP grants apply and where users can switch between grants.[1][11]
- **The Claude team plan for scientists and the AI for Science program** (the team plan launched and the AI for Science program expanded on August 27, 2026) give principal investigators free or discounted seats and researchers API credits of up to $50,000 per project. The same announcement said researchers in biology and chemistry "will still be limited to our Opus-class models" and pointed to the government-partnered access program then in development.[8]

The announcement quotes three companies: [Xaira Therapeutics](https://aiwiki.ai/wiki/xaira_therapeutics), [Edison Scientific](https://aiwiki.ai/wiki/edison_scientific) and Manifold Bio. Manifold Bio said it welcomed "Anthropic's approach of pairing access with accountability."[1] Anthropic said it expected "to enroll hundreds of organizations within the first week, and to scale the program further to support the majority of the life science community in the coming weeks."[1]

## Threat intelligence context

The LSVP announcement links its design to Anthropic's September 2026 threat intelligence report, which it says shows "increasingly sophisticated misuse attempts happening on our platform, including attempts that could support biological weapons development."[1] The report covers activity disrupted between December 2025 and August 2026 and presents five case studies of actors using Claude in ways that could support biological weapons work. These include a grant application for gain-of-function research on chikungunya virus, a research plan involving highly pathogenic avian influenza, orthopoxvirus research, and two cases involving venoms and toxins.[9] The report says none of its misuse cases, across all harm areas, involved Fable or Mythos-class models, apart from one illicit distillation case.[9]

In the toxin cases, the report says the work "proceeded largely unimpeded by our biological safety classifier," by design, and concludes that "a classifier cannot simultaneously enable benefit and prevent harm" in highly technical dual-use areas. It adds that "the only safe way to serve frontier biological capabilities is to offer them in trusted user programs."[9] In its conclusions, Anthropic wrote that safe access "will necessarily require account and institutional signals to verify user legitimacy, and the rudimentary observability provided by data retention to identify misuse."[9] The same report notes that some threat groups try "to gain access to restricted models via AI vendors, evaluators, and trusted access programs."[9]

## Reception

Endpoints News covered the September 17 opening under the headline "Anthropic starts allowing some life sciences researchers to use Mythos."[23] Writing for MIXED, Shane S. Ellison called the High-risk Use grant "the unusual one." He noted that vetting an organization does not stop any of the three threat models Anthropic names, "which is the case Anthropic makes for monitoring usage against each stated use case." He also wrote that Anthropic's version "goes a step further for biology" than tiered access elsewhere in the industry, because for an approved project "none of the safeguards that block life science requests apply."[24]

The program opened a week after CNN and Axios covered Anthropic's threat report. Axios reported "increasing calls for the government to review emerging AI models," and quoted Tom Inglesby, director of the Johns Hopkins Center for Health Security: "The public needs to know most powerful models being developed in this country are being reviewed for high-end national security risks."[25][26] Anthropic's LSVP materials describe US government involvement in developing the program and in widening High-risk access to Mythos, but do not name the agency involved.[1][4]

## See also

- [Claude Mythos 5.1](https://aiwiki.ai/wiki/claude_mythos_5_1)
- [Claude Fable 5.1](https://aiwiki.ai/wiki/claude_fable_5_1)
- [Cyber Verification Program](https://aiwiki.ai/wiki/anthropic_cyber_verification_program)
- [Claude Science](https://aiwiki.ai/wiki/claude_science)
- [Claude for Life Sciences](https://aiwiki.ai/wiki/claude_for_life_sciences)
- [Constitutional Classifiers](https://aiwiki.ai/wiki/constitutional_classifiers)
- [Responsible Scaling Policy](https://aiwiki.ai/wiki/responsible_scaling_policy)
- [AI safety](https://aiwiki.ai/wiki/ai_safety)

## References

1. [Introducing the Life Sciences Verification Program](https://www.anthropic.com/news/life-sciences-verification-program) - Anthropic, September 17, 2026.
2. [What is the Life Sciences Verification Program?](https://support.claude.com/en/articles/16975617-what-is-the-life-sciences-verification-program) - Claude Help Center, accessed September 29, 2026.
3. [Life Sciences Verification Program](https://claude.com/form/life-sciences-verification-program) - Claude by Anthropic (interest form page), accessed September 29, 2026.
4. [Introducing Claude Fable 5.1 and Claude Mythos 5.1](https://www.anthropic.com/claude-fable-and-mythos-5-1) - Anthropic, September 1, 2026.
5. [Claude Mythos](https://www.anthropic.com/claude/mythos) - Anthropic product page, accessed September 29, 2026.
6. [System Card: Claude Fable 5.1 & Claude Mythos 5.1](https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card) - Anthropic, September 1, 2026.
7. [Improving Fable 5's biology safeguards](https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards) - Anthropic, August 7, 2026.
8. [Expanding our support for scientists](https://www.anthropic.com/news/expanding-support-for-scientists) - Anthropic, August 27, 2026.
9. [Detecting and countering misuse of AI: September 2026](https://www.anthropic.com/threat-intelligence-report-september-2026) - Anthropic Threat Intelligence, September 2026. Section "Detecting and countering biological misuse of AI".
10. [Developing Enterprise Frontier Safeguards with our customers](https://www.anthropic.com/news/enterprise-frontier-safeguards) - Anthropic, September 1, 2026.
11. [Safeguards](https://claude.com/docs/claude-science/safeguards) - Claude Science documentation, accessed September 29, 2026.
12. [Claude Science changelog](https://claude.com/docs/claude-science/changelog) - Claude Science documentation, entries for versions 0.1.49 (September 16, 2026) and 0.1.51 (September 22, 2026).
13. [Introducing Claude Opus 5.5](https://www.anthropic.com/claude-opus-5-5) - Anthropic, September 22, 2026.
14. [System Card: Claude Opus 5.5](https://anthropic.com/claude-opus-5-5-system-card) - Anthropic, September 22, 2026.
15. [Introducing Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) - Anthropic, September 28, 2026.
16. [System Card: Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5-system-card) - Anthropic, September 28, 2026. Section 2.2.4.
17. [Why Claude switched models in your conversation with Sonnet 5.5](https://support.claude.com/en/articles/17161993-why-claude-switched-models-in-your-conversation-with-sonnet-5-5) - Claude Help Center, accessed September 29, 2026.
18. [Why Claude switched models in your conversation with Opus 5 or Opus 5.5](https://support.claude.com/en/articles/16049681-why-claude-switched-models-in-your-conversation-with-opus-5-or-opus-5-5) - Claude Help Center, accessed September 29, 2026.
19. [Why Claude switched models in your conversation with Fable 5 or Fable 5.1](https://support.claude.com/en/articles/15363606-why-claude-switched-models-in-your-conversation-with-fable-5-or-fable-5-1) - Claude Help Center, accessed September 29, 2026.
20. [Data retention practices for Covered Models](https://support.claude.com/en/articles/15425996-data-retention-practices-for-covered-models) - Claude Help Center, accessed September 29, 2026.
21. [Anthropic (@AnthropicAI) post announcing LSVP applications](https://x.com/AnthropicAI/status/2100646837799834096) - X, September 17, 2026.
22. [Anthropic doubles a science benchmark score with Fable 5.1 while OpenAI says its Astra model crosses critical cyber threshold](https://www.rdworldonline.com/anthropic-doubles-a-science-benchmark-score-with-fable-5-1-while-openai-says-its-astra-models-crosses-critical-cyber-threshold/) - R&D World (Brian Buntz), September 1, 2026.
23. [Anthropic starts allowing some life sciences researchers to use Mythos](https://endpoints.news/anthropic-starts-verification-program-for-using-mythos-in-biology/) - Endpoints News (Andrew Dunn), September 18, 2026.
24. [Anthropic loosens Claude's biology safeguards for vetted teams, removes them for some projects](https://mixed-news.com/en/anthropic-loosens-claude-biology-safeguards-vetted-teams/) - MIXED (Shane S. Ellison), September 19, 2026.
25. [Anthropic says it blocked possible attempts to use AI to develop bioweapons](https://www.cnn.com/2026/09/10/health/anthropic-bioweapons-report) - CNN (Deidre McPhillips), September 10, 2026.
26. [How AI makes biological research more dangerous](https://www.axios.com/2026/09/11/ai-warnings-biological-research-dangerous) - Axios, September 11, 2026.
27. [Claude Science](https://claude.com/docs/claude-science/overview) - Claude Science documentation, accessed September 29, 2026.
28. [Claude for Life Sciences](https://www.anthropic.com/news/claude-for-life-sciences) - Anthropic, October 20, 2025.
29. [Claude Fable 5 and Claude Mythos 5](https://www.anthropic.com/news/claude-fable-5-mythos-5) - Anthropic, June 9, 2026.