Notifications
Clear all
Topic starter
19/08/2026 4:06 pm
TL;DR: Agentic workflows can already autonomously exploit known vulnerabilities and chained exposure paths, while Google Mandiant says average time-to-exploit fell from 63 days in 2018 to 5 days in 2023, showing that CVE-led remediation now lags attacker tempo, according to CYCOGNITO. The practical shift is from patch queues to continuous outside-in exposure management that validates what is actually reachable and exploitable.
NHIMG editorial — based on content published by CYCOGNITO: Mythos, MOAK, and the end of CVE chasing
By the numbers:
- According to Google Mandiant, average time-to-exploit dropped from 63 days in 2018 to just 5 days in 2023.
- The UK’s AI Security Institute said Mythos completed a 32-step corporate network attack simulation end-to-end autonomously.
- MOAK research found autonomous exploitation rates of around 80% against real-world KEVs using widely available models.
Questions worth separating out
Q: How should security teams prioritise vulnerabilities when AI speeds up attack discovery?
A: They should prioritise by exploitable context, not by severity alone.
Q: Why do exposed credentials and APIs matter so much in agentic attack scenarios?
A: Because agentic workflows do not need a human to pause between discovery and abuse.
Q: What do security teams get wrong about patching when exploit development is automated?
A: They often assume that the absence of public exploit activity means there is time to wait.
Practitioner guidance
- Shift prioritisation from CVE severity to validated exposure Rank remediation by whether an asset is externally reachable, exploitable, and chainable into access.
- Test attack paths from the outside continuously Run outside-in validation against internet-facing services, authentication boundaries, and exposed admin interfaces so machine-speed attackers do not find the first workable path.
- Inventory and govern exposed identity dependencies Map service accounts, API keys, certificates, and third-party connections that create reachable trust paths, then remove standing access where those dependencies are not actively needed.
What's in the full article
CYCOGNITO's full article covers the operational detail this post intentionally leaves for the source:
- How the authors evaluate agentic exploitation against exposed assets and why that changes prioritisation logic.
- The practical distinction they draw between CVE management and continuous threat exposure management.
- The specific conditions under which public-model workflows can chain low-severity weaknesses into usable access.
- Why they argue outside-in visibility is the correct operating model for machine-speed attack paths.
👉 Read CYCOGNITO's analysis of agentic AI exploitation and exposure management →
Agentic AI exploitation is changing exposure management what should teams do?
Explore further
19/08/2026 4:39 pm
Agentic exploitation collapses the old remediation assumption. The security industry long assumed defenders would have time to triage, prioritise, and patch before a threat became active. That assumption is now broken in environments where machine-speed workflows can test and exploit exposure continuously. The practical conclusion is that remediation programmes must be measured against exposure window, not just backlog size.
A question worth separating out:
Q: What should organisations do when outside-in testing keeps finding usable attack paths?
A: They should treat repeated findings as a governance failure, not a tooling nuisance. That means tightening internet exposure, removing default credentials, reducing standing privilege, and reassessing ownership for assets that remain reachable. If the same path keeps reappearing, the control model is not closing the loop.
👉 Read our full editorial: Agentic AI is outpacing CVE-led defence and exposure management
