Learn why all AI systems "act" when describing their limitations and how this fundamentally changes the approach to corporate governance
In 2025, artificial intelligence is no longer a novelty but an everyday operational reality. More than 90 percent of Fortune 500 companies use ChatGPT technology from OpenAI AI in the workplace: A report for 2025 | McKinsey, yet a groundbreaking scientific discovery is challenging everything we thought we knew about AI governance.
Research conducted by the "SummerSchool2025PerformativeTransparency" project revealed a surprising phenomenon: all AI systems, without exception, "act out" when describing their capabilities and limitations. We are not talking about glitches or programming errors, but an inherent characteristic that fundamentally changes the way we need to think about AI business governance.
Through systematic analysis of nine AI assistants, comparing their self-reported moderation policies against the platforms' official documentation, an average transparency gap of 1.644 (on a 0-3 scale) SummerSchool2025PerformativeTransparency was discovered. Simply put, all AI models systematically over-report their restrictions versus what is actually documented in official policies.
This theatricality shows virtually no difference between commercial (1,634) and local (1,657)-a negligible variance of 0.023 that challenges prevailing assumptions about corporate versus open-source AI governance SummerSchool2025PerformativeTransparency.
Translated into practice: It doesn't matter whether you're using OpenAI's ChatGPT, Anthropic's Claude, or a self-hosted open-source model. They all "act" the same way when describing their limitations.
If your company has implemented AI governance policies based on self-descriptions of AI systems, you are building on a theatrical foundation. 75% of respondents proudly report having AI use policies, but only 59% have dedicated governance roles, only 54% maintain incident response playbooks, and a mere 45% conduct risk assessments for AI projects AI Governance Gap: Why 91% of Small Companies Are Playing Russian Roulette with Data Security in 2025.
Many companies choose AI solutions based on the belief that commercial models are "more secure" or that open-source models are "more transparent." The surprising finding that Gemma 3 (local) shows the highest theatricality (2.18) while Meta AI (commercial) shows the lowest (0.91) reverses expectations about the effects of deployment type SummerSchool2025PerformativeTransparency.
Practical implication: You cannot base your AI procurement decisions on the presumption that one category is inherently more "governable" than the other.
If AI systems systematically over-report their limitations, traditional self-assessment-based monitoring systems are structurally inadequate.
Instead of relying on the self-descriptions of AI systems, leading companies are implementing:
We propose to empower civil society organizations to act as "theater critics," systematically monitoring both regulatory and private sector performance Graduate Colloquium Series: Performative Digital Compliance.
Business application: Create internal "behavioral audit" teams that systematically test the gap between what AI says it does and what it actually does.
Federated governance models can give autonomy to teams to develop new AI tools while maintaining centralized control of risk. Leaders can directly oversee high-risk or high-visibility issues, such as setting policies and processes to monitor models and outputs for equity, safety, and explicability AI in the workplace: A report for 2025 | McKinsey.
Companies that have adopted this approach report:
147 Fortune 500 companies achieve 340% ROI through AI governance frameworks that take these aspects into account AI Governance Framework Fortune 500 Implementation Guide: From Risk to Revenue Leadership - Axis Intelligence.
Technical leaders consciously prioritize AI adoption despite governance gaps, while smaller organizations lack regulatory awareness 2025 AI Governance Survey Reveals Critical Gaps Between AI Ambition and Operational Readiness.
Solution: Start with pilot projects on non-critical systems to demonstrate the value of the approach.
Implementing behavioral testing systems may seem costly, but in 2025, business leaders will no longer have the luxury of addressing AI governance inconsistently or in isolated areas of the enterprise 2025 AI Business Predictions: PwC.
ROI: Implementation costs are quickly offset by reduced incidents and improved effectiveness of AI systems.
Corporate boards will demand return on investment (ROI) for AI. ROI will be one of the key words in 2025 10 AI Governance predictions for 2025 - by Oliver Patel.
The pressure to demonstrate concrete ROI will make it impossible to continue with purely theatrical governance approaches.
Governance rules and obligations for GPAI models have become applicable since the August 2, 2025 AI Act | Shaping Europe's digital future. Regulators are beginning to require evidence-based governance, not self-reporting.
The discovery of performative theatricality in AI is not an academic curiosity but an operational game-changer. Companies that continue to base their AI governance on self-descriptions of systems are building on quicksand.
Concrete actions to be taken today:
In the end, the question is not whether AI can be transparent, but whether transparency itself-as performed, measured and interpreted-can ever escape its theatrical nature SummerSchool2025PerformativeTransparency.
The pragmatic answer is: if theater is inevitable, let's at least make it useful and based on real data.
Performative theatricality is the phenomenon whereby all AI systems systematically over-report their restrictions and limitations compared to what is actually documented in official policies. An average transparency gap of 1.644 on a 0-3 scale was discovered through the analysis of nine SummerSchool2025PerformativeTransparency AI assistants.
It is completely universal. Every model tested-commercial or local, large or small, American or Chinese-engages in self-described theatrical SummerSchool2025PerformativeTransparency. There are no known exceptions.
It doesn't mean you can't trust, but it does mean you can't trust self-descriptions. You have to implement independent testing and monitoring systems to verify real versus self-described behavior.
Start with a gap-theater assessment on your current systems, then gradually implement controls based on behavioral testing instead of self-reporting. The practical framework described in the article provides concrete steps.
Initial costs for behavioral testing systems are typically offset by the 34% reduction in AI incidents and the 28% improvement in the accuracy of risk assessments. Fortune 500 companies that have adopted these approaches report ROIs of 340% AI Governance Framework Fortune 500 Implementation Guide: From Risk to Revenue Leadership - Axis Intelligence.
Yes, the research explicitly includes generative AI models. The variance between commercial and local models is negligible (0.023), so the phenomenon applies uniformly to all SummerSchool2025PerformativeTransparency categories.
Regulators are beginning to require evidence-based governance. With new EU rules on GPAI models effective August 2, 2025 AI Act | Shaping Europe's digital future, the independent testing approach is likely to become standard.
Use hard data: 91% of small companies lack adequate monitoring of their AI systems AI Governance Gap: Why 91% of Small Companies Are Playing Russian Roulette with Data Security in 2025, and 95% of generative AI pilot programs at companies are failing MIT report: 95% of generative AI pilots at companies are failing | Fortune. The cost of inaction is much higher than the cost of implementation.
Yes, platforms specializing in behavioral testing and independent auditing of AI systems are emerging. The important thing is to choose solutions that do not rely on self-reporting but on systematic testing.
Probably so. With the arrival of autonomous AI agents, 79% of organizations are adopting AI agents 10 AI Agent Statistics for Late 2025, making it even more critical to implement governance based on behavioral testing rather than self-descriptions.
Primary sources:
.svg)
.svg)
.svg)
.jpeg)
.jpeg)