Anthropic has verified Lawyerd for cyber defence. Here is what it changes for clone takedowns
Reviewed by Ihor Makushinsky Senior counsel · since 2014 A clone casino is a shop window. The damage happens behind it: the “official app” that sends a player’s login to someone else, the deposit form that pays a stranger’s wallet, the kit that puts the same site back up on a new domain the morning after a takedown. Most brand protection removes the window. The machine behind it keeps running, and the clone comes back.
This week Anthropic approved Lawyerd for Defense Access in its Cyber Verification Program (CVP). To our knowledge, Lawyerd is the first brand protection company in the program. Our investigators now work with Claude models whose cyber safeguards are set for verified defensive work. That includes Claude Mythos 5.1, a model Anthropic offers only through its trusted access programs. In plain terms, we can now take the machine apart instead of only reporting the window.
What the Cyber Verification Program is
Anthropic’s generally available models carry conservative cyber safeguards. They help with code review and alert triage. Anthropic’s own help centre says that other work, such as malware analysis, may be interrupted by its safety classifiers. That is deliberate: the skill that dissects a phishing kit can also help someone build one.
The CVP is how verified defenders get past that wall. On 6 October 2026 Anthropic expanded it into three tiers:
| Tier | Who it is for | What it adds |
|---|---|---|
| Defense Access | Security teams defending what they own or maintain, smaller security firms, researchers | Incident response, reverse-engineering malware, analysing and validating vulnerabilities |
| Red Team Access | Organisations doing authorised penetration testing | Adversarial testing of systems they are authorised to test |
| Specialized Access | A small set of organisations, reviewed with the US government | Testing safety systems such as power grids and interbank infrastructure |
Before approving anyone, Anthropic verifies the organisation and the person applying, reviews a description of their security work, and asks them to attest to the security controls each tier requires. For Defense Access that means multi-factor authentication now and phishing-resistant MFA by 15 December 2026.
What Defense Access allows, and what it does not
Defense Access is for defence, and Anthropic has tested that. In its CyScenarioBench evaluation of multi-stage offensive operations, the Defense Access tier blocked Claude Opus 5.5 in 46 of 50 trials. We would not want it any other way. We take attacks down. We do not run them.
What it opens up is the work that clone investigations kept running into: unpacking a malicious app, reading obfuscated scripts, identifying a phishing kit and finding where it sends what it steals.
Every tier includes Claude Mythos 5.1. Anthropic describes it as the same model as Fable 5.1 with safeguards built for cybersecurity, and as the strongest cyber model it has released. Until the October expansion, Mythos-class access was limited to a set of US organisations. Anthropic reports that the partners who had it through Project Glasswing found at least 129,000 verified software vulnerabilities between April and July 2026.
Why a takedown practice needs malware analysis
Our work is removing clone casinos, look-alike domains, fake apps and brand impersonation under named statute. Three things change once we can read what the clone actually does.
The fake app names its own backend. A fake “official” APK has to send stolen logins and card details somewhere. That address is inside the app: a server, an admin panel, a Telegram bot. Taking down the download page removes one link. Reporting the backend to its host cuts off every copy of the app at once.
The kit gives away its siblings. Clone sites are rarely built by hand. They run on kits, and a kit leaves fingerprints: the same scripts, the same panel paths, the same payment endpoints. Identify the kit once and you can find every domain running it. Then you file against all of them together, at the registrar and the host, before the next mirror goes live.
Proof of harm moves faster than proof of copying. A copyright notice says “this site copied our design”. A technical report says “this site steals credentials and intercepts deposits”, and hosts and registrars treat that as urgent abuse. The copied design is still the legal ground. The analysis is the evidence that gets a host to act today instead of next week.
The first system we point it at is our own: the client cabinet and the evidence store that hold every matter.
What does not change
A model reads the code. A lawyer decides what is filed and signs it. Every notice still goes out under a named legal authority (DMCA §512, trademark, Berne, GDPR Article 17) after human review. Our MCP server works the same way: the agent does the groundwork, and the rights holder’s signature stays with a person. CVP access is used for defensive work only, under Anthropic’s Usage Policy.
If a clone of your brand keeps coming back after every takedown, send it to us. We will show you what is behind it.
Asked before engagement.
What is Anthropic’s Cyber Verification Program?
An access program from Anthropic that gives verified security professionals Claude models with reduced cyber safeguards. Since 6 October 2026 it has three tiers (Defense Access, Red Team Access and Specialized Access), each with its own verification and security requirements. Every tier includes Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1.
What does Defense Access allow?
Defensive security work: security operations and incident response, reverse-engineering malware, and analysing and validating vulnerabilities. Offensive work stays blocked. In Anthropic’s own testing, the tier blocked 46 of 50 multi-stage offensive scenarios.
Is Lawyerd an Anthropic partner?
No. The Cyber Verification Program is an access program, not a partnership or an endorsement. Anthropic verified who we are and what security work we do, and granted access on that basis.
Is Lawyerd the first brand protection company in the program?
To our knowledge, yes. As of November 2026, no other brand protection company has announced Cyber Verification Program access. Anthropic does not publish a list of participants, which is why we say "to our knowledge".
Does an AI file takedown notices for Lawyerd?
No. Models analyse the clone, the app and the kit. Counsel decides what is filed and signs it, under a named statute.
What goes into the model?
Attacker artefacts: the clone’s pages and scripts, the fake app, the phishing kit. Analysing a clone does not need access to your systems or your customer data.
Can our own security team apply?
Yes, if the work is defensive. An organisation applies once through Anthropic’s Verification Portal and its admins assign seats. Independent researchers can apply for Defense Access on a paid plan. Anthropic aims to decide on Defense Access within days. Red Team Access takes a few weeks.
Senior counsel · in IP and compliance practice since 2014. Every guide is reviewed before publication.
Full counsel profile →The 48-hour confidential audit.
Send the article, the clone or the URLs and a brief note — counsel reviews every intake personally and responds within one business day. NDA-first; correspondence from dmca@lawyerd.net.
Send us the clone that keeps coming back