Summary of Major Developments
- Announced June 9 at the UK Government’s AI Adoption Summit: BT Group CEO Allison Kirkby confirmed the partnership during her opening keynote at the UK Government’s AI Adoption Summit on June 9, 2026. BT has joined Anthropic’s Project Glasswing — the cybersecurity initiative that grants controlled access to Claude Mythos Preview, Anthropic’s most powerful AI model for vulnerability hunting — making BT the first UK company to publicly confirm Project Glasswing membership. BT stops four million cyberattacks across its networks every day.
- Access to Claude Mythos Preview for vulnerability scanning across BT’s networks and customer infrastructure: The partnership grants BT access to Claude Mythos Preview to strengthen cybersecurity defences across its own networks and the managed security services it provides to customers. Claude Mythos Preview — the original Project Glasswing model that shocked the global cybersecurity community when it was revealed to have found thousands of high-severity vulnerabilities across every major operating system and web browser — will be used by BT to identify systemic vulnerabilities in BT’s infrastructure before malicious actors can exploit them. BT stated its participation reflects its position as a component of the UK’s critical national infrastructure and its role as a provider of managed security services to enterprises.
- Kirkby commits to sovereign British AI capability development: At the AI Adoption Summit, Kirkby made a commitment beyond the BT-Anthropic partnership itself: ‘AI only works at scale when it is underpinned by future-ready networks that are secure, resilient, safe.’ She committed to ‘working with Government to support the further development and deployment of sovereign British AI capability, so that the UK can be an AI maker and not just a taker.’ The statement positions BT as an active participant in UK AI strategy rather than a passive adopter — and commits Anthropic’s most powerful cybersecurity AI to UK national infrastructure defence.
Technical Breakdown: What Claude Mythos Preview Does for BT
Project Glasswing was launched by Anthropic in April 2026 in direct response to the cybersecurity risk posed by AI-powered offensive tools. The core thesis of Project Glasswing is that AI models have reached a level of coding capability where they can surpass all but the most skilled human security researchers at finding and exploiting software vulnerabilities. If AI tools of this capability are used offensively before defenders have equivalent tools, the asymmetry between attack and defence widens catastrophically. Glasswing deploys the same capability to defenders — critical infrastructure operators, government cybersecurity agencies, and vetted security researchers — before it becomes widely available to attackers.
Claude Mythos Preview’s demonstrated vulnerability-finding capability is the foundation of its value to BT. At its April 2026 launch, Anthropic disclosed that Mythos Preview had already found thousands of high-severity vulnerabilities spanning every major operating system and web browser — including decades-old vulnerabilities in code that was considered highly secure and had been examined by thousands of human security researchers without discovery. The implication for BT’s infrastructure is direct: BT operates networks, systems, and software stacks that have been maintained and periodically audited by human security teams for years. Claude Mythos Preview can scan those same systems at machine speed, with a pattern-recognition capability that covers the full space of known vulnerability classes rather than the focused attention of a human audit.
The operational model for BT’s Glasswing deployment — as described by Anthropic and confirmed by BT’s stated data governance requirements — involves strict separation between Claude Mythos Preview and raw customer data. Mythos Preview analyses network telemetry and code, not customer personal data. BT maintains enterprise-grade data governance boundaries that prevent the model from accessing customer data even as it analyses infrastructure logs and vulnerability patterns. This is the security enclave model that Anthropic provides to enterprise Glasswing partners: the model operates within defined data access boundaries, processes anonymised infrastructure telemetry, and returns vulnerability findings rather than raw data.
The scale of the threat BT faces makes the Glasswing partnership commercially urgent. BT’s own data indicates it blocks four million cyberattacks per day across its networks. Human security operations centres cannot manually parse the telemetry volume generated by four million daily attack attempts — the data rate is simply too high for human analysis to be the primary detection mechanism. Large language models like Claude Mythos Preview provide a mechanism to process incident logs, correlate attack patterns, and identify new threat signatures faster than traditional heuristic security systems, at a scale that matches the attack volume rather than the human analyst capacity.
| Project Glasswing Expansion Timeline | Date | Scale | Scope |
| Initial launch — founding partners | April 2026 | Select partners only | AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks + others |
| Wave 2 expansion | June 2, 2026 | ~150 new organisations across 15+ countries | Power, water, healthcare, communications, hardware vendors — Australia, Canada, France, Germany, Italy, Switzerland, Netherlands, Spain, Belgium, Sweden, India, Japan, New Zealand, South Korea |
| BT partnership confirmed | June 9, 2026 | First UK company confirmed | UK critical national infrastructure — telecom networks and managed security services |
| Claude Fable 5 public release | June 9, 2026 | All public API users | Consumer version with cybersecurity classifiers — reroutes to Opus 4.8 for offensive security queries |
| Trusted access programme expansion | Forthcoming | Further Glasswing expansion | Mythos 5 — same capability with some classifiers removed — for approved Glasswing partners |
Commercial and Enterprise Market Impact
The BT-Anthropic partnership is commercially significant at two levels. At the UK national infrastructure level, it establishes a precedent for how the UK’s critical infrastructure operators should be engaging with frontier AI cybersecurity capability — not as a theoretical future application but as a live operational deployment against real daily attack volumes. Ofcom’s April 2026 open letter to telecoms providers — calling on them to ‘assess security risks arising from frontier AI models and take appropriate mitigating action’ — provides the regulatory context that made BT’s Glasswing partnership not just commercially optional but strategically necessary.
At the enterprise cybersecurity market level, BT’s partnership is a signal that telecoms operators — who provide managed security services to thousands of enterprise customers — are beginning to deploy AI-native vulnerability assessment capabilities that will eventually flow through to their commercial security products. BT Business already provides AI-powered cybersecurity solutions to customers of all sizes. Methods developed to secure BT’s core infrastructure will be the foundation for the next generation of managed security services it sells to enterprise customers. A BT customer buying managed security services in 2027 will be receiving protection that is informed by Claude Mythos Preview’s analysis of BT’s own network vulnerabilities — a capability that no alternative managed security provider currently offers.
“BT stopping four million attacks per day is the number that reframes this partnership. That is not a theoretical threat model — it is a live operational reality that no human security team can manually analyse at that volume. Claude Mythos Preview is not an AI tool that BT is experimenting with. It is operational infrastructure that matches the scale of the threat BT faces every day. The partnership is not about innovation — it is about survival at network scale.” — UK Telecom Security Analyst, national infrastructure research, June 10, 2026
“Kirkby’s commitment to UK sovereign AI capability at the AI Adoption Summit is the statement that the UK government needed to hear. The UK has been watching the US and China accelerate AI infrastructure investment and wondering where the UK fits. BT committing to partner with the world’s most capable cybersecurity AI model — and simultaneously committing to sovereign British AI development — is the UK’s answer to that question. It positions BT as infrastructure for UK AI sovereignty, not just a telecom operator.” — UK Technology Policy Analyst, national AI strategy research, June 10, 2026
Frequently Asked Questions
What is Project Glasswing and why did BT join?
Project Glasswing is Anthropic’s cybersecurity initiative launched in April 2026, giving controlled access to Claude Mythos Preview — Anthropic’s most powerful AI model for identifying software vulnerabilities — to operators of critical infrastructure, government cybersecurity agencies, and vetted security researchers. BT joined Project Glasswing on June 9, 2026, becoming the first UK company to publicly confirm membership. BT joined because it stops four million cyberattacks per day across its networks and needs AI-scale vulnerability detection to stay ahead of increasingly sophisticated AI-powered attacks. The partnership was announced by BT CEO Allison Kirkby at the UK Government’s AI Adoption Summit.
Is BT using Claude Mythos 5 or Claude Mythos Preview?
BT’s Project Glasswing partnership provides access to Claude Mythos Preview — the original Glasswing model that has found thousands of high-severity vulnerabilities across major operating systems and browsers since its April 2026 deployment. Claude Mythos 5 was released simultaneously with Claude Fable 5 on June 9, 2026, and is available as an upgrade to Mythos Preview for existing Glasswing partners. BT’s partnership announcement on June 9 references access to Claude Mythos Preview; whether BT will also access Mythos 5 through the Glasswing trusted access programme has not been separately confirmed.
How does Claude Mythos Preview access BT’s customer data?
It does not. BT’s deployment model — as described by both BT and Anthropic for Glasswing enterprise deployments — involves strict separation between Claude Mythos Preview and customer personal data. Mythos Preview analyses network telemetry and code within secure enclaves that prevent access to customer data even while analysing infrastructure logs and vulnerability patterns. BT maintains enterprise-grade data governance boundaries, and Anthropic provides the technical framework for secure enclave deployment that enforces these boundaries at the model access level. The model returns vulnerability findings from infrastructure analysis — it does not process or have access to customer communications, personal data, or account information.
Sources
TechRadar. (2026, June 9). AI is changing cyber security fast: BT becomes first UK firm to join Anthropic Project Glasswing. https://www.techradar.com/pro/ai-is-changing-cyber-security-fast-bt-becomes-first-uk-firm-to-join-anthropic-project-glasswing
Total Telecom. (2026, June 9). BT takes shelter under Anthropic’s Glasswing. https://totaltele.com/bt-takes-shelter-under-anthropics-glasswing/
ISPreview UK. (2026, June 9). BT to Harness Claude Mythos AI Preview to Boost UK Telecoms Security. https://www.ispreview.co.uk/index.php/2026/06/bt-to-harness-claude-mythos-ai-preview-to-boost-uk-telecoms-security.html
Security Brief UK. (2026, June 9). BT joins Anthropic cyber project to bolster defences. https://securitybrief.co.uk/story/bt-joins-anthropic-cyber-project-to-bolster-defences
Telecoms Tech News. (2026). BT bolsters network security with Anthropic Claude Mythos AI. https://www.telecomstechnews.com/news/bt-network-security-with-anthropic-claude-mythos-ai/
Cybersecurity Dive. (2026, June 2). Anthropic shares Mythos with 150 more organizations. https://www.cybersecuritydive.com/news/ai-anthropic-claude-mythos-project-glasswing-expand/821714/
Anthropic. (2026, April). Project Glasswing launch announcement. https://www.anthropic.com/news/project-glasswing
