Security Posture Review
A posture review pays off when it ranks findings by what attackers are exploiting and what your AI systems can reach, not by how many controls are missing. Even strong teams fix only a fraction of exposures quickly, so the order matters more than the length of the list.
Part of Security, Assurance & Governance, in the Think mode. 7 sources, checked .
- of breaches began with exploitation of a vulnerability, now the top entry point (Verizon DBIR 2026)
- 31%
- of breaches began with exploitation of a vulnerability, now the top entry point (Verizon DBIR 2026)
- Verizon Business, 2026(opens in a new tab)
- of CISA Known Exploited Vulnerabilities fully remediated by organizations in 2025, down from 38%
- 26%
- of CISA Known Exploited Vulnerabilities fully remediated by organizations in 2025, down from 38%
- Verizon Business, 2026(opens in a new tab)
- global average cost of a data breach, a record high (IBM Cost of a Data Breach 2026)
- $4.99m
- global average cost of a data breach, a record high (IBM Cost of a Data Breach 2026)
- IBM Newsroom (Cost of a Data Breach Report 2026, conducted by Ponemon Institute), 2026(opens in a new tab)
The problem
Attackers are getting in faster than most teams patch. Verizon's 2026 DBIR found that exploitation of vulnerabilities is now the most common way into a breach, at 31%, ahead of credential abuse1. Only 26% of vulnerabilities in CISA's Known Exploited Vulnerabilities catalog were fully remediated in 2025, down from 38% the year before, and the median time to full fix rose to 43 days1. IBM puts the global average cost of a breach at a record $4.99 million in its 2026 study2.
AI systems add surface that older reviews miss. Among breaches that targeted AI models or applications in IBM's 2026 study, the most common causes were weaknesses in surrounding systems: compromised APIs, applications or plug-ins (27%) and cloud misconfigurations affecting AI workloads (27%)2. Third parties widen it further. Verizon found a third party involved in 48% of breaches, and only 23% of third-party organizations fully fixed missing or weak MFA on their cloud accounts1. A review that stops at the corporate network will not see the model endpoint, the vector store or the vendor's tenant.
Boards and regulators now ask for evidence. SEC rules require an Item 1.05 Form 8-K, generally within four business days of deciding an incident is material, and a description of how you assess and manage material cyber risk and how the board oversees it3. The final phase of NYDFS Part 500 took effect on 1 November 2025, with MFA for all individuals accessing information systems and a documented asset inventory4. NIST CSF 2.0 added a Govern function, and NIST's draft Cyber AI Profile maps securing AI system components onto CSF 2.0 outcomes5,6.
Good looks like a short, ranked list. Verizon's data shows that organizations at their best fix only 30% to 40% of KEV instances in the first week after detection, which makes choosing the right ones the real strategy1. A useful review scores each finding by known exploitation, exposure and what an attacker could reach through it, including AI agents and service accounts. It hands your teams a sequence they can execute and your board a summary it can rely on.
What you get
- 01
AI and data asset inventory
A map of the models, endpoints, agents, vector stores, pipelines and service accounts in scope, with an owner for each and the data each one can reach.
- 02
Control assessment against NIST CSF 2.0
A scored view of the Govern, Identify, Protect, Detect, Respond and Recover outcomes as they apply to your AI and data platforms, with the evidence behind each score.
- 03
Exposure testing
Configuration review of the cloud accounts and AI workloads in scope, a check of exposed vulnerabilities against the CISA KEV catalog, and targeted tests of AI interfaces for prompt injection and data leakage.
- 04
Third-party review
How the vendors that host or process your AI data authenticate, segment and log, including MFA on their cloud accounts, checked rather than taken from a questionnaire.
- 05
Ranked remediation plan and board summary
Findings ordered by exploitation, exposure and reach, each with an owner, plus a governance summary written for the people who sign disclosures and certifications.
Where it goes wrong
The failure modes we design against from the first week, and the control for each.
A long list with no order
Reports that weigh every gap equally swamp teams. In Verizon's data, even the best organizations close only 30% to 40% of KEV instances in the first week1. The control: rank findings by known exploitation, internet exposure and reach, using the KEV catalog as an input to prioritization as CISA recommends7.
Leaving AI out of scope
In IBM's 2026 study, breaches aimed at AI most often came through APIs, plug-ins and misconfigured cloud workloads around the model2. A review scoped to the classic network misses them. The control: put AI assets, their identities and their data flows in the inventory before testing starts.
Ignoring the vendor's side
A third party was involved in 48% of breaches in the 2026 DBIR1. Questionnaires rarely show whether a vendor enforces MFA or logs access to your data. The control: technical checks of authentication and logging for any supplier that holds or processes sensitive data.
Findings that never reach the board
Public companies must describe their processes for managing material cyber risk and the board's oversight of it3. A technical report nobody above the CISO reads leaves that disclosure unsupported. The control: a governance section that ties findings to risk appetite and owners, using CSF 2.0's Govern function5.
Standards and rules
NIST Cybersecurity Framework (CSF) 2.0
Primary text on nist.gov(opens in a new tab)Six functions, including the new Govern function, that work as a common scoring frame so security, risk and the board read the same results.
SEC cybersecurity disclosure rules (Form 8-K Item 1.05, Regulation S-K Item 106)
Primary text on sec.gov(opens in a new tab)Public companies disclose material incidents generally within four business days of the materiality decision and describe risk processes and board oversight in the annual report on Form 10-K; the review should produce evidence for both.
NYDFS 23 NYCRR Part 500 (Second Amendment)
Primary text on dfs.ny.gov(opens in a new tab)Since 1 November 2025, covered entities need MFA for all individuals accessing information systems and written procedures for a complete asset inventory; a review should test both directly.
CISA Known Exploited Vulnerabilities catalog
Primary text on cisa.gov(opens in a new tab)CISA's authoritative list of vulnerabilities exploited in the wild, which it says organizations should use as an input to vulnerability prioritization.
How to measure it
- Share of AI assets (models, endpoints, agents, service accounts) in the inventory with a named owner
- Number of KEV-listed vulnerabilities open past your target window
- Median days to fully remediate critical findings
- Share of human and service accounts covered by MFA, with phishing-resistant MFA for privileged access
- Share of critical vendors with verified MFA and logging on systems that hold your data
Questions buyers ask
How is a security posture review different from a penetration test?
A penetration test tries to break in through a defined scope and reports what worked. A posture review looks across governance, identity, configuration, vulnerability management, logging, response and third parties, and tells you which gaps matter most. Most reviews include targeted testing, but the output is a ranked plan for the whole environment rather than a list of exploited paths.
What does a security review of AI systems cover that a standard review misses?
Model endpoints and the APIs around them, plug-ins and tools an agent can call, vector stores and retrieval pipelines, the service accounts AI workloads run under, training and fine-tuning data flows, and the model providers you send data to. It also tests AI-specific attacks such as prompt injection and sensitive data leaking through model output.
Which framework should a posture review be scored against?
Pick the one your board and examiners already use. NIST CSF 2.0 is a common choice because it covers governance as well as technical controls and maps to most other standards. Regulated firms then add their sector rules on top, for example NYDFS Part 500, the HIPAA Security Rule or NERC CIP, so findings line up with what auditors will test.
Sources
Every figure on this page comes from the sources below. We checked each link on .
- 1.2026 Data Breach Investigations Report(opens in a new tab)Verizon Business, May 19, 2026 · verizon.com
- 2.IBM Study: One in Four Malicious Breaches are AI-Enabled, Costing Companies $6 Million on Average(opens in a new tab)IBM Newsroom (Cost of a Data Breach Report 2026, conducted by Ponemon Institute), July 29, 2026 · newsroom.ibm.com
- 3.SEC Adopts Rules on Cybersecurity Risk Management, Strategy, Governance, and Incident Disclosure by Public Companies(opens in a new tab)U.S. Securities and Exchange Commission, July 26, 2023 · sec.gov
- 4.Cybersecurity Implementation Timeline for Covered Entities (23 NYCRR Part 500)(opens in a new tab)New York State Department of Financial Services, November 2023 · dfs.ny.gov
- 5.NIST Releases Version 2.0 of Landmark Cybersecurity Framework(opens in a new tab)National Institute of Standards and Technology (NIST), February 26, 2024 · nist.gov
- 6.NIST IR 8596 (Initial Preliminary Draft): Cybersecurity Framework Profile for Artificial Intelligence (Cyber AI Profile)(opens in a new tab)National Institute of Standards and Technology (NIST), December 16, 2025 · csrc.nist.gov
- 7.Known Exploited Vulnerabilities Catalog(opens in a new tab)Cybersecurity and Infrastructure Security Agency (CISA) · cisa.gov