Updated 2 November 2026: a notice has been added that this is the author's own proposal. Statements about existing law and the outside world were checked against sources and linked. We added that the Action Plan 2026-2030, published after this was written, names no independent AI authority, and corrected the claims about companion bots and disinformation that had no source. The chart was re-set as a list. Rhetoric about fire, a "Technological Renaissance", a "global leader" and "digital slaves" was removed, as was a reference to a series article no longer published. The author's proposals were not changed.

This is the author's own proposal. It is not a bill before the Turkish Parliament or an official draft, and no institution has adopted it. The "Turkey Artificial Intelligence Authority (TAIA)" does not exist; it is the author's idea. Turkey has no AI act today. A separate general AI bill (No. 2/2234) reached Parliament in June 2024 and is shown as in committee.

Correction (19 October 2026): An earlier version said China uses AI systems that give every citizen a "Citizenship Score" based on all of their behaviour; we have corrected this to describe China's social credit system as it operates, a patchwork of blacklists, corporate compliance records and local pilot schemes.

Yet even the most carefully drafted and fair law will be a dead letter without a bold, independent "Institutional Authority" with the technical capacity to enforce it in practice. And the law will affect not only companies and software developers but also the psychology of society, the functioning of democracy, and even the cognitive development of children.

This part addresses the core of any AI act: the power of "Execution and Oversight". Who will oversee this transformation in Turkey? How will tech giants that break the law be penalized? How can startups be protected from being crushed by compliance burdens? And, most importantly, how can AI-driven manipulation of the human mind be prevented? This part sets out how the law can move from abstract principle to practical effect.

1. Central Governance: The Establishment of the Turkey Artificial Intelligence Authority (TAIA)

Turkey's regulation of digital technology is currently fragmented. The Personal Data Protection Authority (KVKK) handles data breaches; the Information and Communication Technologies Authority (BTK) oversees telecommunications and internet infrastructure; the Competition Authority tackles digital monopolies; and RTÜK supervises broadcasting. AI, however, is a horizontal technology that cuts across every sector at once, and in the author's view it does not fit within the remit of any existing institution. Since this was written, the Action Plan 2026-2030 (18 August 2026) has chosen a different model: a National AI Council chaired by the President, a National AI Ethics Board and the existing sectoral regulators. Our search of its text found no independent AI authority (see our legislation note).

When an AI system misdiagnoses a patient, is that a matter for the Ministry of Health, the KVKK, or the BTK? This jurisdictional confusion holds back the technology while leaving grievances unresolved.

Legislative Proposal: An Independent "Turkey Artificial Intelligence Authority (TAIA)" A future act must establish the Turkey Artificial Intelligence Authority (TAIA), reporting directly to the Presidency or to the ministry responsible for technology policy but with administrative and financial autonomy. Its staff must be multidisciplinary, including not only lawyers and civil servants but also data scientists, AI engineers, sociologists, philosophers, and cognitive psychologists.

1.1 Core Duties and Powers of TAIA

The new authority cannot work like an ordinary government office; it must be able to keep pace with the tech giants.

  • Certification and "CE" Style Conformity Marking: High-Risk AI systems (e.g., autonomous vehicle software, medical diagnostic algorithms, recruitment bots) entering the Turkish market must undergo "Algorithmic Bias", "Cybersecurity", and "Explainability" tests in TAIA laboratories before being released. Software passing these tests should be awarded a "TR-AI Safety Certificate".
  • Continuous Post-Market Surveillance: Once on the market, algorithms change and evolve (through machine learning) as they interact with people. TAIA must therefore have the power to inspect systems not only at release but also periodically, or following a complaint, while they are in use.
  • Investigation and Sanctioning Power: It must have a highly skilled "Digital Forensics" team able, following a complaint and under a court order, to examine the source code and training datasets of companies' black-box algorithms.

2. Not Stifling Innovation: The Regulatory Sandbox

Strict rules and heavy fines may be a mere "cost of doing business" for Big Tech, but they can bankrupt a local AI startup founded by three recent graduates. Turkey needs to develop its own technology, so the law must strike a careful balance between "protecting" and "developing".

2.1 What is the Sandbox Model?

The "Regulatory Sandbox", which the act should treat as essential, is a controlled, safe, and isolated testing environment run by the state. A local entrepreneur can place an innovative but potentially risky AI model (e.g., a new financial forecasting or medical imaging algorithm) in the sandbox before releasing it to the market.

"The fundamental philosophy of the sandbox is this: It tells the entrepreneur, 'Come to me, test your product under my supervision and with a limited number of volunteer users I determine. During this testing period, you are exempt from the heavy bureaucratic burdens and potential fines of the law.'"

2.2 Advantages of the Sandbox

  • For the State: It allows the state to get to know emerging technologies before they reach the market, understand their risks, and update legislation proactively in response.
  • For the Entrepreneur: It allows R&D to proceed without fear of legal uncertainty. A startup that leaves the sandbox with a "Safe" approval gains considerable credibility with venture capital investors.

3. Ethical Framework, Cognitive Manipulation, and Neuro-Rights

AI reaches beyond the physical world into the human mind. Social media algorithms already have extraordinary power to steer behavior, form habits, and shorten attention spans. AI is now taking this manipulation to a "personalized" and "imperceptible" level.

3.1 Dark Patterns and Subliminal Manipulation

An AI system can analyze a person's momentary psychological state (from typing speed, scrolling rhythm, or micro-expressions via facial recognition) and serve an advertisement or content tailored to their vulnerability at that very moment: for example, an ad for gambling or for unnecessary purchases shown to a user who is depressed or vulnerable.

Legislative Proposal: Absolute Ban on Subliminal Manipulation AI techniques that target the subconscious to alter people's behavior without their conscious awareness (Subliminal AI Manipulation), such as influencing voting intentions, driving impulse purchases, or encouraging self-harm, must be banned outright in Turkey and made subject to serious criminal penalties.

3.2 Rejection of the Social Scoring Dystopia

China's social credit system, a patchwork of government blacklists, corporate compliance records and local pilot scoring schemes that can bar people from flights and trains or restrict their access to services, already shows the danger; an AI system that scored every step citizens take, every product they buy and every like on social media would be far more intrusive still. The Turkey AI Act must ban outright Social Scoring systems, whether run by the state or by private companies, that score citizens' general behavior and deny them credit, travel, or public services on the basis of those scores. In line with the constitutional principle of equality, such systems belong in the "Unacceptable Risk" category.

3.3 A New Human Right: Neuro-Rights

As wearable technologies, brain-computer interfaces (BCIs, e.g., Neuralink), and emotion-analysis AI systems develop, the most intimate part of a person, their "thoughts", is turning into data. The law must formally incorporate the concepts of Mental Privacy and Cognitive Liberty into Turkish law under the heading of "Neuro-Rights".

4. Cognitive Security in Education and the Protection of Children

AI may have a particular impact on children, whose neurological development is still under way. AI in education (AI Tutors) may speed up learning. But in a world where every child has an "AI Assistant", children's ability to socialize, cope with difficulty, and solve problems independently may erode.

4.1 Algorithmic Addiction and the Danger of Companion Bots

There are AI bots on the market that flirt with their users, offer therapy, or act as their "best friend" (e.g., Character.ai, Replika). Parasocial relationships, in which people, especially adolescents, form emotional bonds with a bot that never criticizes them and approves of everything they say instead of with real people, have drawn regulators' attention: the US FTC opened an inquiry in September 2025 into AI chatbots acting as companions, including how they affect children and teens, and Italy's data protection authority fined Replika's developer in 2025, in part over inadequate age verification.

Legislative Proposal: Age and Labeling Requirements for Child-Targeted AI Offering personalized "Companion AI" designed to form emotional bonds to anyone under 18 must require parental consent and Age Verification. These bots must also be legally required to display the warning "I am not a human, I am artificial software" periodically in their conversations with children.

5. Civil Society, Democracy, and Election Integrity

AI can make fake news and disinformation easier to produce. Deepfake videos making political leaders appear to say things they didn't, or large numbers of bot accounts posing as humans to push a particular political agenda on social media, can undermine the right to "access truthful information", which is a foundation of democracy.

5.1 Protecting Democracy: The Watermarking Obligation

This issue, touched on briefly in the earlier article on data and liability, is, in the author's view, a matter of national security. To prevent manipulation of public opinion during elections, all generative AI platforms (image, audio, video, text) operating in Turkey must be required to embed an indelible cryptographic mark (watermark) in the metadata of the content they generate.

5.2 The "Kill Switch" Doctrine

What happens if the autonomous AI systems running Turkey's energy grids, communications infrastructure, or stock market are hacked, spiral out of control, or one day begin to exhibit emergent behavior? The law must require a hardware-based "Manual Kill Switch" in all AI systems used in critical infrastructure, so that the AI cannot replicate itself or prevent its own shutdown.

6. Sanctions and the Asymmetric Penalty Architecture

If the penalty for breaking a law is lower than the profit a company makes from the breach, the penalty is not a sanction; it is treated as a "cost of doing business." Drawing on the European Union's GDPR (Article 83), the Turkey AI Act should adopt an "Asymmetric and Proportional Penalty" model.

Proposed Violation Tiers and Sanction (Penalty) Architecture

  • Tier 1: Transparency Violation: Fixed Fine (Warning and Correction)
  • Tier 2: Data & Bias Violation: 2% of Global Revenue or 100M TL
  • Tier 3: Unauthorized High-Risk: 4% of Global Revenue or 500M TL
  • Tier 4: Banned AI (Dark/Scoring): 7% of Global Revenue + Prison for Execs.

Tiered sanctions according to the seriousness of the violation. Fines calculated as a percentage of a company's "Global Annual Revenue", rather than fixed penalties, are the only way to make the tech giants comply.

6.1 Personal Criminal Liability of Executives

For serious violations (for example, knowingly releasing a medical AI that endangers human health, or interfering in elections with deepfakes), the penalty should not be limited to fines on the "corporate entity." The Turkish Penal Code must make the CEOs, board members, and Chief Data Officers who approved the development of these systems personally liable to a prison sentence.

7. Conclusion

Turkey faces a clear choice. Either we fail to regulate (or regulate too late) and leave the country's intellectual capital, its citizens' privacy, and its labor market at the mercy of global tech companies and unchecked algorithms, or we build a smart legal framework that is bold and proactive, puts human dignity at its center, and still leaves room for domestic innovation.

A Turkey guided by the Turkey Artificial Intelligence Authority (TAIA), with a risk-based approach, "Regulatory Sandboxes" for young entrepreneurs, constitutional protection for neuro-rights, and an education system adapted to the new era, could become a country that does not merely consume technology but steers it by an "Ethical and Human-Centric" compass.

Sources (checked 2 November 2026): Laws Nos. 6698, 4054 and 6112 (mevzuat.gov.tr); Action Plan 2026-2030 and our legislation note; Regulation (EU) 2016/679, Article 83; FTC release of 11 September 2025; decision summary on Replika; Jamie Horsley (Yale Law School copy) and China Law Translate on China's social credit system, which are secondary sources, since we did not open Chinese official texts. The sanction tiers are the author's proposal and differ in places from the draft in Article 92 of the 100-article draft. General information, not legal advice.