Every exam cycle generates a wave of misconceptions. The SEC's priorities documents are concise, leading readers to fill in gaps with assumptions. This year, these assumptions pose significant risks for CISOs, as the Division of Examinations (EXAMS) has included artificial intelligence, referred to as "super intelligence" or SI, in areas traditionally seen as outside the security domain: fiduciary conduct, conflicts of interest, and third-party oversight.
The priorities were published on October 23, 2026. Here's where common interpretations go astray.
Myth 1: AI is solely a cybersecurity issue
Reality: EXAMS addresses AI under emerging technologies, describing its use in investment analysis, customer management, code development, fraud detection, back-office operations, AML and Reg BI compliance, and trading. This encompasses nearly every control function in a firm.
Examiners will scrutinize policies for monitoring AI use across these functions. A security-only governance model can't address whether an AI-assisted recommendation tool aligns with a client's profile. Under the COSO ERM Framework, AI risk falls within governance and culture, strategy, and performance, not just information security. Assign a business owner to each AI use case, and have security manage the control environment around it.
Myth 2: AI exposure is primarily about external threats
Reality: While threat actors using AI to enhance attacks is a concern, EXAMS expects firms to prepare for AI-related cyberattacks and ensure identity theft and incident response programs account for AI-generated threats. Phishing and other scams are specific examples.
However, your own AI deployment is equally important. EXAMS requires firms to detail their AI inventory, governance, and monitoring controls. An organization that can't list the AI tools its staff and vendors use faces an inventory gap before a threat gap. Shadow AI, or tools adopted without review, is a potential exam finding.
Myth 3: Economic conflicts are solely a compliance issue
Reality: EXAMS will prioritize reviews of advisers' fiduciary obligations related to economic conflicts, as highlighted in its June 9, 2026, Risk Alert. Examples include fee rebates, fee calculations, cash management, and quid pro quo arrangements.
Consider a scenario where an adviser uses a model that ranks products for recommendations, and the vendor's data or logic favors revenue-sharing products. Detecting such conflicts requires model documentation, change control, and access to outputs, which are governed by a CISO's program. AI can also mitigate conflict risk by identifying anomalies in fee calculations. The duty of loyalty now extends through your systems.
Myth 4: Outsourcing to a vendor transfers the risk
Reality: EXAMS will focus on advisers using third-party service providers, including those accessing client accounts, and on advisers relying on proxy advisory firms. Examiners will assess how advisers evaluate proxy recommendations and whether reliance meets legal obligations.
This logic applies to AI vendors. If a third party's model interacts with client data, your firm remains accountable. Practical diligence includes:
- Contractual rights to audit or obtain assurance reports on AI-related controls
- Clarity on data processing locations and model training
- Incident notification terms for AI-enabled attacks
- Documented evaluation of vendor outputs
Reliance without evaluation is a pattern EXAMS will test.
Myth 5: Dropped priorities can be deprioritized
Reality: Items from the previous year's list, like Rule 605 disclosures and the Regulation SHO bona fide market-making exception, are absent but not irrelevant. Obligations persist, and examiners often address issues outside the published list.
The same caution applies to your AI program. Don't build controls solely based on one year's document. For example, cryptoassets aren't a priority for a second year, yet tokenization is mentioned. Treat priorities as directional signals and build programs around underlying obligations: standards of care, information security, and operational resiliency.
What to Do Instead
Replace myth-driven planning with an evidence-based program:
- Build and maintain an AI inventory. Include internally developed tools, vendor features, and unsanctioned use. Record purpose, data accessed, owner, and risk tier.
- Document governance and monitoring. Examiners will ask for descriptions of both. Define approval gates, ongoing testing, and escalation paths, and keep records.
- Update incident response and identity programs for AI-generated attacks. Include scenarios like synthetic-voice or deepfake-assisted social engineering in exercises.
- Map AI use cases to conflict risk. Work with compliance to identify where models influence recommendations or valuations and test for embedded conflicts.
- Tighten third-party oversight. Extend diligence to AI features within existing vendors, especially those with client account access.
- Report upward. Provide the board or risk committee with a clear view of AI exposure, as accountability doesn't end with the CISO.
The 2027 Priorities don't question whether you use AI. They ask if you can demonstrate awareness of where it is, what it does, and who is responsible. Build that evidence now, before an examiner requests it.





