Policy Pulse - Issue #40 | Week of November 1, 2026
The UK promises researchers a Computer Misuse Act defence, but only for the accredited ones. The DMCA security research exemption heads for renewal, and Anthropic tells the SEC that AI bug hunting is a legal risk.
Policy Pulse - Issue #40 | Week of November 1, 2026
Your weekly briefing on cybersecurity policy affecting vulnerability disclosure and security research.
In 60 Seconds
- UK Home Office: a Computer Misuse Act defence is coming, but only for "accredited" researchers, and in a different bill; the Lords never voted on the review amendment.
- US Copyright Office: the DMCA 1201 security research exemption is headed for renewal with no opposition, and none of the ten new proposed classes covers security or AI research.
- Anthropic: its IPO filing tells investors that researchers have used its models to find and exploit flaws in other companies' systems, sometimes inside authorized bug bounties.
- CISA: a new strategy that never mentions CVE, KEV or coordinated disclosure, and ten KEV additions in five days, most with three-day deadlines.
- Next deadline: Monday November 2, Lords Third Reading of the Cyber Security and Resilience Bill, and NIST IR 8623 comments close.
Top Story
The UK will give researchers a Computer Misuse Act defence. It will only cover the ones it accredits.
Amendment 158 to the Cyber Security and Resilience Bill would have required ministers to review, within 12 months, a statutory defence under section 1 of the Computer Misuse Act for good-faith researchers. At Report stage on October 26 it never reached a vote. Lord Clement-Jones spoke to it while moving Amendment 61, withdrew that, and when 158 came up in its own place Hansard records: "Amendments 158 to 162 not moved." (Hansard)
The reason was the government's answer. The minister told the House: "The Home Office has already carefully reviewed the Computer Misuse Act and proposes to introduce a defence to Section 1 for accredited cyber security researchers when carrying out certain cyber security activity that would currently be unlawful under that section." That defence will arrive in the national security bill promised in May's King's Speech, "with measures to update the Computer Misuse Act. This legislation is in development." Clement-Jones closed: "We will look forward to that Bill when it comes forward with that provision in it." The same evening peers rejected Amendment 41, which would have required frontier AI developers to report unauthorised access to essential services, starting within 24 hours, 95 to 165. Third Reading is Monday. (Hansard, VinciWorks)
Why it matters for VDP: After a decade of CyberUp campaigning, the UK has committed on the record to a defence. The word that matters now is "accredited." No scheme exists yet, and whoever designs it decides whether an independent researcher, a student, or a researcher outside the UK who reports to a UK program is covered. The fight moves from whether to who, and it happens when the national security bill is published.
Throwback: In Issue #39, we flagged 158 as the possible first UK vote on a researcher defence. There was no vote, and a promise instead.
Upcoming Deadlines & Events
| Date | What | Who should act | What to do |
|---|---|---|---|
| Nov 2 | UK House of Lords: Third Reading, Cyber Security and Resilience Bill; then back to the Commons | UK researchers, policy folks | Follow the bill page and brief your MP before Commons consideration |
| Nov 2 | NIST IR 8623: CSF 2.0 profile for federal O-RAN deployment, comments close | 5G and O-RAN researchers | Comment via NIST CSRC |
| Nov 10 | KEV due date for Brotli CVE-2026-111497 ("Node2Shell"), added Oct 27 | Federal operators, vendors shipping Brotli | Check every bundled copy against the KEV catalog |
| Nov 20 | UK ICO: agentic AI call for evidence closes, with a Data Security section | VDP operators, researchers | Respond via the ICO |
| Nov 30 | NIST SP 800-82 Rev. 4: OT security guide, comments close | ICS and OT researchers | Comment via NIST CSRC |
| Dec 4 | NIST SP 800-100 Rev. 1: Information Security Handbook for executives, rebuilt around CSF 2.0, comments close | Program owners who brief executives | Comment via NIST CSRC |
| Dec 7 | US Copyright Office: first-round comments on the 1201 proposed classes; same day, NIST SP 800-221 Rev. 1 comments close and agencies must adopt BOD 26-04 timelines | Researchers, program operators | File via the 1201 rulemaking page; comment on SP 800-221r1 |
| Dec 11 | CISA 2015 information-sharing liability shield and the stopgap funding bill both expire | Policy folks | Track the reauthorization via Nextgov |
| Dec 31 | UN Convention against Cybercrime: signature closes (96 signatories, 3 parties) | Policy folks | Watch the UN Treaty Collection for the year-end push |
This Week in Policy
AI & Emerging Tech Security
Anthropic tells investors that AI-assisted bug hunting is a legal risk, including inside bug bounties. Anthropic's SEC filing ahead of its IPO says the legal and regulatory risk "extends to independent security researchers and others who have used, and may in the future use, our models to identify and exploit vulnerabilities in the systems of other companies," at times in connection with authorized bug bounty programs, and that "There can be no assurance, however, that the safeguards or access regimes we adopt will prove adequate." The filing also discloses Commerce export controls on Mythos 5 and Fable 5 and an FTC civil investigative demand. (CyberScoop)
Why it matters for VDP: A frontier lab has now told the SEC that AI-assisted research, even inside an authorized program, is a legal risk. If your safe-harbor language is silent on AI-assisted and agentic testing, that silence is now a documented ambiguity. Say whether it is in scope.
Japan's AI Safety Institute: an open-weight model reached the top exploit tier on eight tasks. In a new evaluation note, Japan AISI reran GLM-5.3 and GLM-5.2 on exploit development for known V8 vulnerabilities after revising its settings. T1 is the top tier, and arbitrary code execution is one of the capabilities assessed there: "GLM-5.2 reached T1 on none of the tasks, Claude Opus 4.8 on one, and GLM-5.3 on eight." Its point is that average scores hide this: "Even on the same tasks, the capabilities demonstrated differed substantially between models." (Japan AISI)
Why it matters for VDP: CAISI called GLM-5.3 "the most cyber-capable open-weight model released to date" in September; Japan now finds exploit-tier capability in downloadable weights that no access program governs. Expect working proofs of concept from reporters no access program answers for.
Federal Strategy & Regulation
CISA's new strategy leaves out CVE, KEV and disclosure, and the person who ran them still has no successor. "CISA 2030: Defend, Harden, Prevail," released October 26, promises that "Within 60 days, CISA will begin reviewing its core services and capabilities." The 2026 strategy document never mentions the CVE program, KEV, coordinated disclosure or researchers; its software line is that CISA "will coordinate with partners to develop data-driven guidance and best practices to drive software security and supply chain transparency." (CISA, strategy PDF) A week after Nextgov reported the departure of vulnerability response chief Lindsey Cerkovnik, Risky Bulletin noted: "CISA has not announced a replacement." (Risky Bulletin)
Why it matters for VDP: A 60-day review of "core services" with no named owner for CVE and KEV is the moment to tell CISA what the disclosure ecosystem depends on. Nobody else is going to make that case in the review.
CVE & Vulnerability Programs
The Brotli fix sat in a public repo for eight months; KEV now gives most bugs three days. The fix for CVE-2026-111497 "was committed to google/brotli on 27 February 2026, with no release and no advisory." Fixed releases arrived in October, after the bug had spread into Node.js, .NET, Bun, Envoy and Python HTTP clients. CISA added it to KEV on October 27, due November 10. (Help Net Security) It was one of ten KEV additions between October 26 and 30, alongside both PHP2Shell CVEs and GitLab CVE-2026-97229 (CVSS 10.0, due November 1), which an attacker had already tried against Alpine Linux's own GitLab: "The attack came close to working." By our count of CISA's KEV feed, 29 of October's 36 additions carried deadlines of three days or less under BOD 26-04. (CISA KEV, The Hacker News)
Why it matters for VDP: A silent public commit is a disclosure to anyone who diffs repos, and the federal clock after exploitation is now shorter than most vendor triage SLAs. Advisory timing and quality upstream decide how much of those three days is left downstream.
Tailscale patches flaws published about two hours after it got the report. Erica Windisch of Cyberdione Labs published "Tailworm," which she describes as "reported but not coordinated," about two hours after Tailscale received it on October 24. Tailscale shipped 1.104.4 with four bulletins (TS-2026-014 to 017) on October 26, two of them in Tailscale SSH, and none carries a CVE ID. (The Hacker News, Tailscale)
Why it matters for VDP: The vendor's two-day fix made an uncoordinated drop survivable. The missing CVE IDs make it invisible to every scanner and KEV workflow keyed on them.
Legal & Researcher Protections
The DMCA 1201 security research exemption is on track for renewal; nothing new for AI research. The Copyright Office's notice of proposed rulemaking, published October 28, says of the good-faith security research exemption that "No comments were filed opposing renewal," and that the conditions behind it "are likely to continue during the next triennial period." Professor Michael Specter's renewal petition says security researchers "play a vital role." The Office consolidated 25 petitions into ten proposed classes: six audiovisual, jailbreaking, repair, and two for video games. None covers security research or AI trustworthiness research. First-round comments are due December 7. (Copyright Office, Federal Register)
Why it matters for VDP: The safe harbor researchers already have looks set through 2030. AI red-teaming will have to wait for the next cycle, so the evidence for it needs collecting now.
International Developments
The CRA reporting platform is live, but researchers can't use it yet. ENISA's FAQ says the Single Reporting Platform, live since September 11, "has not yet implemented the voluntary reporting functionality provided for under Art. 15(1) and (2) of the CRA." Article 15 is the route a third-party finder would use. (ENISA)
Why it matters for VDP: Manufacturers' mandatory reports flow; a researcher's voluntary report has no route in yet, so the CSIRT you report to is the whole path.
Worth Reading
- Monitor AI agents' actions in real time (Partnership on AI, MIT Technology Review): "Right now, we're effectively flying blind when deploying these tools." Agent telemetry is the evidence trail that will separate authorized agentic testing from an incident.
Policy Pulse is a weekly bulletin from disclose.io. Have a tip or an experience that belongs in the next issue? Join the community discussion.
