Illinois AI audit law exposes gaps in oversight standards

The gist
Illinois just became the first state to demand annual third-party audits for frontier AI developers, but its bold new law exposes big gaps in how—and by whom—AI safety checks will actually happen.
What to know
- The Artificial Intelligence Safety Measures Act (SB 315) kicks in July 2026 and requires independent audits for AI companies with over 10²⁶ operations or $500M+ revenue, starting in 2028.
- Illinois' law sets high bars for auditor independence but offers no formal certification, leaving 'who audits the auditors' a looming question.
- Without clear rules on access and standards, audits risk becoming box-ticking exercises rather than real guardrails against hidden AI risks.
Illinois Sets New AI Bar
Illinois’ AI audit law targets only the most powerful AI developers, signaling a shift toward multi-state regulation and proactive oversight before harms occur.
Illinois has positioned itself as a trailblazer in AI governance with the enactment of the Artificial Intelligence Safety Measures Act (SB 315) on July 6, 2026, mandating annual independent third-party audits of large frontier AI developers starting in 2028. This pioneering requirement, unique among states, targets a narrow class of companies—those conducting over 10²⁶ computing operations or generating more than $500 million in annual revenue—ensuring oversight of only the most powerful AI systems. By focusing on safety, transparency, and cybersecurity, Illinois complements the regulatory frameworks established by California and New York, signaling a growing multi-state approach to frontier AI regulation.
While the legislation clearly defines compliance obligations for advanced AI developers, the market for independent AI audits and technical standards remains nascent, presenting implementation challenges. Jennifer Dickey highlights that this evolving landscape complicates the identification of qualified auditors, underscoring the need for developing professional standards alongside the legal mandate. Moreover, the Act’s preventative emphasis encourages companies to conduct due diligence on AI vendors, aiming to deter harmful conduct proactively rather than merely providing remedies after violations occur.
Enforcement of SB 315 is robust, with the Illinois Attorney General granted exclusive civil authority to impose penalties up to $1 million for initial violations and $3 million for subsequent offenses. This stringent enforcement framework reflects Illinois’ serious regulatory stance and commitment to holding frontier AI developers accountable, reinforcing the state’s leadership role in shaping responsible AI governance at a critical juncture in the technology’s evolution.
The Missing Auditors Dilemma
With no formal certification or accreditation, Illinois’ law exposes a vacuum in AI audit expertise and independence, forcing the creation of an entirely new assurance profession.
Illinois’ Artificial Intelligence Safety Measures Act mandates annual independent AI audits starting in 2028 but leaves auditor qualifications broadly defined without establishing any formal certification, licensing, or registry system. The law requires auditors to adhere to generally accepted auditing standards and demonstrate competence, including experience with frontier model safety experts, while maintaining strict independence by prohibiting financial ties with the developers they audit. However, this minimalistic framework creates ambiguity around who qualifies as an auditor and how their expertise and impartiality are verified, as there is currently no governing body to admit, accredit, or discipline practitioners in this emerging field.
The absence of an established AI auditing profession presents a significant challenge in identifying qualified auditors who can simultaneously meet the law’s technical and independence criteria. Large accounting firms possess robust auditing methodologies and can recruit AI technical experts, yet they struggle to integrate frontier AI model expertise into assurance frameworks that satisfy both independence and generally accepted auditing standards. Conversely, specialized AI evaluation labs boast deep technical knowledge but often maintain research relationships with AI developers, disqualifying them under the statute’s independence requirement and limiting their role to model testing rather than compliance attestation.
Illinois’ pioneering AI audit mandate underscores the urgent need to develop a new assurance profession tailored to AI safety, encompassing standardized methodologies, evidence protocols, independence rules, peer review mechanisms, accreditation processes, and accountability structures. This evolution mirrors historical precedents where emerging technologies initially lacked formal auditing professions until standards and qualifications were codified—such as the formation of Underwriters Laboratories in 1894 and auditing professions following the 1933 and 1934 Securities Acts. Illinois’ law thus represents an early but critical step toward professionalizing AI governance and establishing a credible assurance ecosystem.
Beyond Checklists: Real AI Oversight
Without deep, independent access and enforceable consequences, AI audits risk becoming safety theater—leaving hidden risks unchecked and oversight vulnerable to industry capture.
The inherent complexity and evolving nature of general-purpose AI systems challenge traditional regulatory frameworks, which often rely on fixed boundaries and checklist-style audits. As TechSpective highlights, AI risks emerge contextually and through unforeseen combinations, requiring governance approaches that can imagine near-infinite use cases across diverse domains, user intents, and cultural contexts. This dynamic landscape demands that Illinois’ AI Safety Measures Act and similar laws evolve continuously to remain relevant, underscoring the urgent need for professionalization of AI auditing practices that go beyond static compliance checks.
Meaningful AI governance hinges on granting independent auditors deep, white-box access to AI models—including full model weights, training data, and development records—to detect hidden risks such as ‘sleeper agent’ behaviors that evade superficial black-box tests. Despite growing consensus among lawmakers and frontier labs like OpenAI, Anthropic, and Google DeepMind on the necessity of independent evaluations, current laws remain vague about what constitutes 'reasonably necessary' access, risking audits devolving into mere safety-washing. Emerging technologies like confidential computing offer promising solutions to balance proprietary concerns with the need for thorough inspections, but legal frameworks must explicitly mandate calibrated access levels tied to model risk and standardized reporting to ensure transparency and accountability.
To establish trustworthy AI oversight, Illinois and other jurisdictions must build robust professional assurance frameworks and independent evaluation bodies insulated from political and corporate influence. Proposals such as the FRONTIER Act’s licensed Independent Verification Organizations (IVOs) emphasize the importance of statutory authority, pooled funding, and protections against conflicts of interest to safeguard auditor independence. Moreover, audit findings should carry enforceable consequences akin to aviation safety inspections, rather than mere recommendations, ensuring that governance mechanisms are proactive, contemporaneous, and capable of preventing incidents like OpenAI’s mid-2026 sandbox escape. Without embedding these structural safeguards and real-time record-keeping into AI governance, oversight risks remaining reactive and superficial.
Beyond catastrophic risks, AI governance must also address subtler, pervasive harms such as eroded human judgment, invisible exclusion, and institutional dependency on AI intermediaries, which affect everyday users in administrative, emotional, and economic spheres. As TechSpective notes, most people will encounter AI not as a dramatic threat but through these nuanced impacts, necessitating audit frameworks that capture these diffuse harms rather than focusing solely on headline-grabbing failures. This broader perspective calls for independent evaluation bodies capable of holistic assessments that consider AI’s societal integration and evolving risks over time, reinforcing the need for continuous professionalization and adaptive governance.

