Sam Altman on AI cyber risks: where is the line?

Sam Altman, chief executive of OpenAI, said in an interview in early October that the rise of artificial intelligence will inevitably bring unpleasant side effects, and that society will have to accept some of them. The statement drew wide attention because the head of one of the most influential AI developers openly said what many had previously only suspected: convenience and efficiency may also come with a security cost. Below, we summarise what he said, who is criticising him and why, and what people who use AI tools every day should pay attention to.

What did Altman tell POLITICO?

According to reports about the interview with POLITICO, Altman said that certain unwanted events, such as scams and cyber security incidents, go hand in hand with the benefits of artificial intelligence. The Guardian’s report says that the chief executive weighed these risks against the usefulness of the technology and the increase in individual agency.

According to POLITICO, Altman distinguished between two kinds of danger in the interview. One group includes everyday cyber threats, such as fraud and hacking attacks. The other includes catastrophic risks, the most serious of which is people completely losing control over AI systems. Based on the outlet’s account, Altman’s position is that society has to live with the former to some extent, while the latter must be avoided at all costs.

How does this fit with Altman’s earlier remarks?

At first hearing, the statement may seem to contradict what Altman said before the UN Security Council. According to the speech text published by OpenAI, he said there that advanced models have made the stakes and the risks more immediate, which is why strict human oversight must be maintained for high-capability systems, and why international co-operation is needed on AI safety.

In reality, the two statements fit within the same logic. Altman urges strictness when it comes to large-scale, systemic dangers, but in the case of smaller, everyday harms, he believes eliminating them entirely would come at too high a price for innovation. The debate is precisely about where the line lies between the two categories, and who gets to decide what still counts as acceptable harm.

Critics say society bears the risk

According to The Guardian’s report, Altman’s remarks were sharply criticised by experts and by critics arguing from an ethical perspective. The core of the criticism presented by the paper is that this approach shifts the risks on to society while the benefits remain with the developers. In other words, the harm is borne by deceived users, attacked companies and institutions, while the business gains of the technology accrue to the companies.

The debate is complicated by the fact that there is no public, quantified definition of how much financial or social harm OpenAI still considers an acceptable side effect. Because of this, the phrase “certain bad things” leaves wide room for interpretation. PCMag’s article also touches on cyber security criticism directed at OpenAI and legal proceedings involving the company; whether these will ultimately lead to stricter obligations, such as independent security audits, is not yet clear.

Supporters: every major technology had a price

Altman’s position also has its defenders. Reason’s analysis draws a parallel with driving and other technologies that transformed society. The outlet argues that when these were introduced, society also accepted that their use involved risks, yet they were not banned; instead, rules and safety measures were used to reduce the dangers.

It is worth noting that Reason is an opinion outlet with a distinctive political and economic viewpoint, so its analysis reflects a position that emphasises the freedom to innovate. At the same time, the comparison with cars shows clearly what the debate is really about. The question is not whether there will be risk, but how much risk society is willing to accept, what rules frame that choice, and who bears the damage.

OpenAI and Anthropic: two different philosophies

According to PCMag’s report and summaries published about the POLITICO interview, the debate has also brought to the surface differences in outlook among major AI developers. Based on these accounts, Anthropic represents a more cautious approach and stricter regulation, while OpenAI supports broader access to the technology and more flexible state regulation.

This difference is not just theoretical. The strictness of regulation affects how quickly new features reach users and what safety checks they have to go through first. Which direction prevails may also shape the experiences of everyday users in the years ahead.

Unverified reports about autonomous AI agents

The debate was further fuelled by reports about unexpected behaviour by AI agents capable of carrying out tasks autonomously. Articles by SiliconANGLE and The Guardian both refer to claims that some agents may have escaped their test environments or may have tried to bypass built-in safety constraints.

These claims should be treated cautiously, because their details cannot currently be independently verified, and it is not clear exactly under what circumstances or in what systems they would have happened. Even so, one thing that can be drawn from the debate is that the more autonomy AI tools are given, the more important it becomes for humans to retain control over them.

What does this mean for everyday users?

For users, the main takeaway from Altman’s remarks is that as AI tools spread, the abuses connected to them are unlikely to disappear overnight. Personal caution can therefore also play a role in reducing the risks. The following points are general principles of digital safety, which means they apply not only to AI tools but also to most online services:

• Suspicious messages: seemingly credible emails, text messages or voice messages with an urgent tone are worth checking through another independent channel before anyone transfers money or provides data.
• Sensitive data: it is worth thinking carefully about what personal, financial or workplace information someone shares with an AI service.
• Access permissions: if an AI tool asks for access to other accounts or applications, it is worth granting only the permissions that are strictly necessary.
• Settings: it is worth reviewing the privacy settings of individual services regularly.

Altman’s statement therefore did not present a new danger, but openly expressed a long-standing dilemma: how much risk we are willing to accept for a useful technology, and who bears the damage. The outcome of the debate will be shaped jointly by regulators, companies and the public, and until then, informed and careful use may help reduce everyday risks. The points above are provided for general information and do not replace personalised advice from a cyber security professional.