# Exposure management: fix first what your tests proved reachable. > Exposure management in ColossalX applies continuous threat exposure management to AI. Exposures are ranked by validated reachability, meaning a control actually missed the attack, and by the business criticality of the agent they hit, not by raw severity. A CISO review rolls up testing programmes and names its gaps, and drift alerts fire when a defence slips. AI exposures ranked by validated reachability and business impact, with drift alerts, a CISO review and a report a person signs. Canonical page: https://colossalx.tech/platform/exposure-management · Last reviewed: 6 Oct 2026 *Illustration:* Validated signals, ranked by reach: coverage gap found calling exposure register; runtime guard found calling exposure register; exploited dependency found calling exposure register; exposure register to fix first (validated); exposure register to your SIEM; exposure register to CISO review (signed report). ## The threat and the control - **The threat:** Hundreds of findings arrive with a severity each, and the one an attacker can actually reach waits. - **The control:** ColossalX ranks first the exposures your own tests proved reachable, on the agents that matter most. ## How it works: From a defence that slipped to a signed report. A technique that used to be blocked on a production agent is missed in a re-validation. It is ranked first, reviewed and reported. ### Workflow: a defence that slipped, ranked and reported (illustrative) 1. **Defence slips.** A scheduled re-validation finds a technique once blocked is now missed. support-bot · ASI01 goal hijack · Blocked -> support-bot · ASI01 goal hijack · Missed | Regression alert 2. **Ranked first.** Validated reachability on a production agent puts it at the top. Exploitability: validated; Agent: support-bot · production; Band: critical 3. **CISO review.** The review reads incomplete if any source could not be read. Red-team runs (done: computed); Live app scans (done: computed); External VAPT (waiting: not tracked here) 4. **Signed and exported.** A person signs the exposure report; ranked exposures go to the SIEM. CISO: Signed off the exposure report | Exported in OCSF | x, tested ## What you see: The CISO review, with its gaps named. One sign-off view over incidents, red-team resilience, live app scans and the AI attack surface, with testing programmes tracked and what needs the CISO's attention. 1. **Collect validated signals.** Coverage gaps, landed attacks, guard firings and exploited dependencies feed the register. Signals come from gaps in the in-path coverage matrix, runtime guard firings, model and dataset file findings, dependency findings carrying known-exploited status and exploit-likelihood scores, and authorisation probe results. The register is only as full as the validation you run. 2. **Rank by reach.** Validated exploitability and business impact, not raw severity, set the order. Each exposure shows its band, source, agent, trust zone and business criticality, and whether its exploitability is validated, detected or theoretical. Evidence can raise an agent's priority, never argue it down. 3. **Watch for drift.** A technique once blocked and now missed raises a regression alert. Coverage is trended per agent across scheduled re-validations. Name and describe a newly published technique, pick an agent, and attacks are synthesised and run in-path the same day, with no product release. 4. **Review, sign, export.** A signed exposure report, and export to your SIEM in OCSF. The report is not final until a person signs it off. Export sends the ranked exposures to your configured webhook or SIEM destinations, or says nothing was exported and why. *Illustration:* An illustrative exposure review: exposures placed by reachability and impact and ranked worst first, each with the agent it reaches and whether it was validated, external testing programmes tracked, and a status that reads incomplete because one scanner could not be reached. ## How we know - Exposures are ranked by validated exploitability and business impact, not by raw CVSS. - The CISO review reads incomplete, not healthy, when a source could not be read. - Programme findings are labelled computed, entered by hand, or not tracked here. - The exposure report is not final until a person signs it off. ## Where a ranked exposure goes next. Ranking decides the order of work; the spine carries it from there. - **Owned work.** A ranked exposure becomes one issue with an owner, closed only on positive evidence. - **The risk register.** Validated exposures write risk entries that resolve when a control stops the attack. - **Re-validation.** Authorised runs re-attack the agent, so a closed exposure is proven, not assumed. - **Your SIEM.** Ranked exposures go to your webhook or SIEM destinations in OCSF, or say why not. Where an x ends up: x, tested. ## Specs: delivery and data - **Delivery:** SaaS, from one login. - **Isolation:** Each customer runs in an isolated workspace with its own database. - **Certifications:** None held. Frameworks are mapped to and assessed against. ## Frameworks - Coverage measured from runs MITRE ATLAS: Technique coverage trended per agent. - Assessed per agent against OWASP Top 10 for Agentic Applications: Gaps ranked by agentic risk, per agent. ## What it does not do - The register is only as full as the validation you run; it stays empty until tests produce signals. - External VAPT, ASV and red-team engagements are tracked as programmes; their findings are not ingested. - Exposure management covers AI agents and the code behind them, not networks or endpoints. - Your SOC alerts are not ingested; detection is read from ColossalX's own gateway and controls. *Illustration:* Assurance programmes · labelled: In-path red team Computed from runs; AEV simulations Computed from runs; Live app scans Computed from scans; External VAPT Findings not tracked here; ASV scans Entered by hand. Each count labelled. ## Questions ### What is continuous threat exposure management (CTEM)? Continuous threat exposure management, or CTEM, is a cycle of finding exposures, ranking them by what an attacker could actually reach and what it would cost, validating them and fixing them in that order. ColossalX applies it to AI agents, ranking exposures by validated reachability and the business criticality of the agent affected. ### How are AI exposures ranked? By validated exploitability and business impact rather than raw severity. Validated means a ColossalX test showed a control actually missed the attack. Business impact comes from the agent: its trust zone and its criticality, which you declare or which is raised when it is seen running in production. Evidence can raise priority, never lower it. ### What does the CISO review show? One sign-off status over open incidents, the latest red-team resilience, live app scans and the AI attack surface, with the share of testing programmes active, the named gaps and repositories under review, worst first. If any source cannot be read, the status reads incomplete instead of healthy. ### Can we track external VAPT, ASV and red-team programmes alongside? Yes, as programmes with an owner, a scope, a cadence, a next due date and a status. Their findings counts are labelled honestly: computed where ColossalX holds the results, entered by hand where a person typed them, or not tracked here for external engagements whose reports never reach the platform. ### Which export formats are supported? Ranked exposures go to your configured webhook or SIEM destinations in OCSF, a standard security-event format, and the exposure report is a branded PDF that is final only after a person signs it off. If no destination is configured, the export says nothing was sent and why. ## Related - [Red-teaming and validation](https://colossalx.tech/platform/red-teaming) - [Threat intelligence](https://colossalx.tech/platform/threat-intelligence) - [Risk quantification](https://colossalx.tech/platform/risk-quantification) --- ColossalX is an AI security and governance platform from Quantexra Labs LLP, delivered as SaaS. Book a walkthrough: https://colossalx.tech/demo · client.success@quantexra.tech