What does Anthropic’s Mythos mean for our AI cyber security risk?
The short answer is this: Mythos itself is unlikely to be turned against your business directly – but its arrival confirms that AI which can hunt for software flaws at machine speed is now a working reality, and that has clear implications for AI cyber security risk over the next eighteen months. For most London businesses, the right response is to treat Mythos as a warning shot and get the basics in order before similar capabilities reach attackers at scale.
What Mythos actually does, in plain English
Mythos is a Claude-based AI tool from Anthropic, designed to find software vulnerabilities at speed across every major operating system and web browser. In the right hands, it’s a force multiplier for security teams. In the wrong hands, it accelerates exactly the same job for attackers. That’s the dual-use problem at the heart of every powerful AI capability – and it’s why the Mythos story matters well beyond Anthropic.
Why AI-assisted vulnerability discovery changes the maths
The gap between a vulnerability being disclosed and being exploited has been shrinking for years. AI-assisted tooling compresses it further. The window to patch, mitigate or detect a flaw is no longer measured comfortably in weeks – in some cases it is hours. Once tools with Mythos-class capability are in the hands of threat actors, that window closes faster still.
The flood of vulnerabilities is already overwhelming the system
The volume of disclosed vulnerabilities is at record levels and rising sharply. The “Common Vulnerability and Exposures” – CVE – is a register of known vulnerabilities maintained by NIST. The programme published 25,059 records in 2022, 28,961 in 2023, 40,077 in 2024 and 48,244 in 2025 – with NIST reporting a 263% increase in CVE submissions between 2020 and 2025. The pressure has been so significant that in April 2026, NIST formally changed how it operates the National Vulnerability Database, moving to a risk-based model in which only CVEs meeting specific criteria are immediately enriched. Everything else is now flagged “lowest priority”. The system defenders rely on to understand which vulnerabilities matter is, by NIST’s own admission, struggling to keep pace.
Why “low severity” no longer means “low priority”
The CVE database assigns “CVSS” scores to vulnerabilities, a 1-10 rating intended to rate vulnerabilities severity. Most organisations triage by CVSS score – patch the criticals, schedule the highs, defer the mediums. AI-assisted tooling like Mythos undermines that habit. Its real power isn’t only finding fresh vulnerabilities; it’s discovering how to chain several lower-severity flaws into a more serious cyber issue. A medium plus a medium plus a misconfiguration can equal an entry route. That kind of analysis used to take a skilled human researcher days. AI does it in minutes.
Attack surface reduction must come first
The first response to a faster threat environment is to give attackers less to work with. Attack Surface Reduction means systematically narrowing the cloud services, AI platforms, applications, infrastructure and endpoints exposed to the outside world – and locking down the configuration of what remains.
In practice it’s less glamorous than it sounds, but it’s where the real wins are. On staff laptops and PCs, it means switching off Windows features the business doesn’t actually use – the kind of legacy capabilities and document-handling shortcuts that attackers love to abuse but that most users never notice are there. On staff sign-ins, it means using Microsoft’s Conditional Access in Entra ID so people can only reach sensitive systems from trusted devices, in expected locations, with strong authentication. On cloud and AI platforms, it means turning off services no one is using, removing forgotten access permissions, and treating any new AI tool with the same scrutiny as a core business system. Every unused service, default setting and over-permissive account is a vulnerability waiting to be chained together.
Vulnerability management – speed is now the differentiator
The second response is to remediate faster, addressing new vulnerabilities swiftly. The metric that matters is mean time to remediation, and improving it requires automation at every step of the pipeline: discovery, prioritisation, patch deployment, verification.
The NCSC’s Cyber Essentials scheme currently requires high and critical vulnerabilities (CVSS 7.0 and above) to be remediated within 14 days of a fix being available – and from April 2026, missing that window is an automatic certification failure. That mandate is already a stretch for many organisations. In a world where AI tooling can chain mediums into criticals and produce working exploits in hours, fourteen days is no longer rapid enough on its own. The bar has to move. Investment in robust, automated vulnerability management pays for itself the first time a critical flaw is closed before exploitation – and gives leaders the assurance that compliance and security are pulling in the same direction.
AI cyber security risk – the bottom line
The honest concern around Mythos isn’t a single tool or a single vendor. It is that the speed of vulnerability discovery and exploitation is accelerating, and the public infrastructure that supports defenders is straining under the load. Organisations that get their house in order now – reducing exposure and speeding up remediation – will absorb that shift. Those that don’t will find the maths working against them.
How we can help
At Doherty Associates, we’ve spent over 34 years helping organisations in highly regulated, risk-sensitive industries strengthen their cyber resilience – with attack surface reduction and automated vulnerability management now forming a core part of that mission.
If you’re reviewing how AI cyber security risk could shape your firm’s defences over the next year, our specialists would be happy to guide you through the best next steps. Learn more on our Managed Cyber Security Services page.
Sources
- CVE Programme, Published CVE Records (Metrics) — cve.org/About/Metrics
- NIST, NIST Updates NVD Operations to Address Record CVE Growth, 15 April 2026 — nist.gov/news-events/news/2026/04/nist-updates-nvd-operations-address-record-cve-growth
- NCSC, Cyber Essentials: Requirements for IT Infrastructure v3.3, April 2026 — ncsc.gov.uk
We’re here to help
If you want to achieve better outcomes for your business through a more intelligent use of technology, talk to us.
Contact us