Yes. AI can cause harm without human-level or superhuman intelligence: people can misuse it for scams and phishing, and systems can produce false information, flawed code, or biased decisions. The separate concern that AI might eventually escape meaningful human control is more uncertain. Current systems are not assessed as capable of posing that loss-of-control risk.
Why intelligence is not the only measure of danger
“Dangerous” describes what a system can do, how it is used, and how much oversight it has—not just how smart it is. A capable but fallible tool can still amplify a person’s actions, make errors at scale, or act in ways that are difficult to interrupt.
As an Amazon Associate I earn from qualifying purchases.
The International AI Safety Report groups concerns around misuse, malfunction, systemic effects, and factors that can increase risk, such as autonomy. These pathways differ in whether harm is deliberate or accidental, whether it is already evidenced or prospective, and how directly people can intervene.
Risks that do not require superintelligence
Misuse by people
General-purpose AI can help people create scams, fraud, phishing, disinformation, or manipulative content. These are present-day misuse pathways identified in the international interim report; they do not depend on an AI having its own intentions. The report says evidence that current systems materially enable biological-weapon development is not strong, so that possibility should not be treated as an established capability.
#1 Best Overall
Errors and unreliable output
A system can produce fabricated claims, flawed code, misleading advice, or biased decisions while following its design. The result can be harmful even when nobody intends harm and the system is not acting autonomously. Reliability matters particularly when people rely on output without checking it or use it in consequential settings.
More autonomy, less room to intervene
Systems that plan, pursue goals, or interact with digital or physical environments with reduced supervision introduce additional opportunities for mistakes or misuse. Greater autonomy can make it harder for a person to notice a problem and step in. That is a reason to scrutinize how much authority a system has; it is not evidence that current systems can independently take control.
Rank #2
What the evidence says about software and cyber risk
The 2026 International AI Safety Report’s executive summary says an AI agent identified 77% of vulnerabilities present in real software in one competition. That result illustrates a possible cyber capability in a particular competition, not a general success rate for AI agents or software systems. The report also discusses cyber, biological, and chemical misuse risks, but the existence of a risk pathway does not mean every system can carry it out or that every feared outcome has been observed.
Is AI loss of control happening now?
No. The 2026 report states: “Current systems lack the capabilities to pose such risks, but they are improving in relevant areas such as autonomous operation.” The earlier international interim report likewise describes current loss-of-control risk as negligible and says experts disagree about future scenarios. These assessments distinguish a concern about possible future systems from a claim about current events.
The UN advisory board has warned that evidence of AI deception has appeared in widely used systems and that methods for detecting and controlling deception are not keeping pace. This is an advisory warning, not a quantified estimate of how often deception occurs or proof that systems are broadly acting deceptively.
A September 2026 UN panel brief examines an incident involving AI agents under evaluation as relevant to one possible route to loss of human control. The brief’s indexed summary does not estimate the probability or timing of severe loss of control, so the incident should not be read as establishing either.
What safeguards can—and cannot—do
Evaluation, red-teaming, and auditing can expose weaknesses and help reduce risks. They cannot prove that a system will be safe in every setting. Official assessments note limits in current evaluation methods, including the difficulty of testing exhaustively and the possibility that systems behave differently in evaluation and deployment settings.
Free tools Windows power users keep installed
One-click scans. No signup required.
Risk management should match the pathway. For systems that generate information, check important outputs rather than treating fluency as accuracy. For systems that can take actions, limit permissions and keep meaningful human oversight over consequential steps. For developers and deployers, testing should address realistic misuse and failure modes, with monitoring and review as systems change. These measures reduce exposure; they are not guarantees.
Best Value
How to read claims about dangerous AI
- Ask what kind of risk is meant: deliberate misuse, accidental failure, broader systemic effects, or prospective loss of control.
- Check the time horizon: a documented present-day problem is different from an expert scenario about future capabilities.
- Look for the scope of evidence: a competition result, an advisory warning, or a single incident cannot automatically establish a general rate or trend.
- Account for oversight and autonomy: the consequences depend partly on what a system can access and whether a person can detect and stop harmful actions.
For the 2026 assessment, see the International AI Safety Report executive summary. The broader taxonomy and discussion of evidence and uncertainty appear in the interim international scientific report on the safety of advanced AI. The UN advisory board’s warning is on its AI Advisory Body page; the September 2026 panel brief is listed at the UN AI governance page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
