AI’s Built‑In Safety Systems Are Hindering Criminals — and Quietly Helping Law Enforcement

Intro: the article below was written by AI under my precise instructions. Videos on YouTube paint a different picture to the one stated in the post. The situation is confused. Interestingly, a huge number of YouTube videos are only getting 1-20 views! Next to nothing. I wonder if YouTube is drowning in videos that are simply not interesting to the public. And/or AI created videos are swamping the website. I think AI will do a lot of harm to YouTube. Fake videos which are excellent in their production are what I am referring to.

---------------------

Recent research suggesting that cybercriminals are struggling to adopt artificial intelligence highlights a broader and increasingly important reality: mainstream AI systems are structurally designed to resist misuse, and this design unintentionally strengthens the position of law enforcement. While AI is not built as a policing tool, its safety architecture makes it far more difficult for criminals to exploit — and that has significant implications for crime prevention and public safety.

At the core of modern AI development is a simple principle: do not enable harm. Major AI providers embed extensive safeguards that prevent models from offering procedural guidance on illegal activities, bypassing security systems, exploiting vulnerabilities, or evading detection. These systems are trained to decline requests that could facilitate wrongdoing, even when the user’s intent is ambiguous. As a result, criminals cannot rely on AI for the kind of detailed, step‑by‑step instructions that would meaningfully enhance their operations.

This refusal behaviour is not accidental. It is the product of deliberate design choices, including filtered training data, reinforcement learning with human feedback, and rule‑based safety layers. These mechanisms ensure that when a user attempts to solicit harmful information, the AI either declines outright or redirects the conversation toward lawful, high‑level explanations. For criminals, this means AI cannot be used as a shortcut to expertise. For law enforcement, it means a powerful potential tool is effectively off the table for those who would misuse it.

Another challenge criminals face is the lack of precision and repeatability. Even when they attempt to disguise their intentions, AI systems avoid providing actionable detail in sensitive areas. Criminal activity often depends on reliable, consistent instructions. AI, by design, introduces uncertainty and vagueness in high‑risk contexts, making it unsuitable for planning or executing illegal operations. This unreliability further reduces AI’s value to criminals.

Moreover, mainstream AI platforms maintain logs, audit trails, and usage monitoring — not for policing, but for safety, quality control, and abuse prevention. Criminals are acutely aware that their interactions may be traceable. This pushes them away from regulated AI systems and toward unregulated, offline, or custom‑built models. Ironically, this migration itself can be informative: when criminals abandon mainstream tools, it reveals the types of capabilities they are seeking and the limitations they face.

The cumulative effect is that AI raises the barrier to entry for criminal activity. Opportunistic offenders who might once have benefited from easy access to technical knowledge now find themselves blocked. More sophisticated criminals must invest in specialised tools, custom models, or human expertise — all of which increase cost, risk, and visibility. In this way, AI functions much like improved locks, stronger authentication, or better surveillance systems: it doesn’t eliminate crime, but it makes it harder, slower, and more detectable.

While AI is not a law‑enforcement instrument, its safety‑first design means it naturally aligns with the goals of crime prevention. By refusing to assist with harmful activity and by limiting the operational value criminals can extract, AI becomes an indirect but meaningful ally in the effort to reduce and contain crime.

--------------------

P.S. please forgive the occasional typo. These articles are written at breakneck speed using Dragon Dictate. I have to prepare them in around 20 mins. Also, sources for news articles are carefully selected but the news is often not independently verified. And, I rely on scientific studies but they are not 100% reliable. Finally, (!) I often express an OPINION on the news. Please share yours in a comment.