Notifications
Clear all
Topic starter
24/07/2026 4:08 pm
TL;DR: AI has collapsed the median time from disclosure to exploitation from about 10 months in 2021 to three hours in 2026, while frontier models are making multi-step attack chains easier to automate, according to Illumio and the U.K. AI Security Institute. The security problem is no longer just patch speed, but how far an attacker can move after the first foothold.
NHIMG editorial — based on content published by Illumio: The Race to Stop AI From Turning Vulnerabilities Into Breaches
By the numbers:
- The median time from disclosure to exploitation has fallen from about 10 months in 2021 to three hours in 2026.
- In a controlled 32-step corporate-network simulation, Mythos Preview succeeded in 3 of 10 attempts and averaged 22 steps.
- Claude Opus 4.6 averaged 16 steps in the same 32-step corporate-network simulation.
Questions worth separating out
Q: What breaks when patching cannot keep up with AI-speed exploitation?
A: Patch-first programmes assume defenders have enough time to validate, approve, and deploy fixes before attackers operationalise a flaw.
Q: Why do weak identity controls make fast vulnerability exploitation worse?
A: Because once an attacker gets a foothold, broad or persistent access lets them convert a single exploit into lateral movement.
Q: How do security teams know whether containment is actually working?
A: They should test whether the identity can still execute privileged actions after revocation, not just whether the API call succeeded.
Practitioner guidance
- Map exploit windows to containment windows Track the time from disclosure to exposure reduction, not just to patch completion, and define the controls that can limit attacker reach during that gap.
- Reduce standing privilege before the next disclosure cycle Review high-risk service accounts, API tokens, and administrative roles for broad or persistent access that could be abused once a foothold exists.
- Build attack-path interruption into response playbooks Add steps that cut lateral movement early, including microsegmentation, privileged session containment, and quarantine logic for compromised workloads.
What's in the full article
Illumio's full blog covers the operational detail this post intentionally leaves for the source:
- A deeper breakdown of how microsegmentation changes attacker reach after the first foothold.
- Examples of visibility and containment controls used to isolate compromised systems in practice.
- The article's specific framing of frontier AI, vulnerability risk, and breach containment for security leaders.
- Context on how Illumio connects AI-speed offense to network and endpoint containment decisions.
👉 Read Illumio's analysis of AI-speed vulnerability exploitation and containment →
AI-speed exploitation and the containment gap teams are missing?
Explore further
24/07/2026 4:12 pm
AI-speed exploitation creates a containment debt problem: the enterprise is no longer only measured by how quickly it patches, but by how much damage an attacker can do before a patch lands. That is a governance shift, not just an operational one. In identity-heavy environments, standing privilege and weak segmentation amplify that debt because the first valid access can still reach too much. Practitioners should treat containment capacity as a board-level resilience metric.
A question worth separating out:
Q: Who is accountable when a sanctioned AI tool causes a data breach?
A: Accountability should sit with the owner of the identity and permissions behind the tool, not only the team that approved the application. If a sanctioned AI workflow can reach sensitive data, the organisation must govern its access path, logging, and containment as rigorously as any other high-risk identity.
👉 Read our full editorial: AI-speed exploitation is collapsing vulnerability response windows
