Sam Altman advocates for widespread AI access but warns of catastrophic risks

In interviews with Politico and Vanity Fair, Sam Altman, CEO of OpenAI, argued that the benefits of artificial intelligence justify accepting certain risks, but not a loss of control over the systems. He presented this position during a conversation on Politico’s Decoded program, which was released earlier this week.

Altman stated that society must “accept some bad things” in exchange for the advancements AI provides and the freedom to use it. He explained that the technology needs to be democratized, allowing people to use the systems and participate in decisions about their future.

Widespread access, with limits to mitigate risks

When discussing the problems that may accompany the spread of AI, the executive mentioned cyberattacks, scams, and misuse. Despite this, he believes the positive impacts will outweigh the negative ones, even without a guarantee of eliminating all abuses.

Altman distinguished these incidents from a scenario in which humanity would no longer be able to control the systems. In such a case, he advocated for caution regarding the speed of progress, warning that accepting known risks does not mean tolerating truly catastrophic threats.

Regulation should focus on the most advanced models

During the interview, Altman argued that rules and tests should be concentrated on frontier models, which are the most advanced. He said he would not support a licensing regime for systems one or two generations behind this frontier, although he acknowledged that they can also cause problems.

When discussing the possibility of the technology becoming concentrated in a single laboratory, the CEO emphasized user autonomy as a priority, describing this scenario as a risk of concentration.

Proposed safety mechanisms

In another interview published by Vanity Fair on Monday, the 5th, Altman explained that his view on regulation “is not a binary issue.” He advocated for combining space for innovation with safeguards for higher-capacity systems.

The mechanisms mentioned include transparency about incidents, learning from accidents, “standards before launch,” and “a liability regime.” He also stated that assessments need to cover the period of model training.

Altman added that the level of security required by OpenAI will need to increase as models become more capable.

Frequently Asked Questions

Does Altman consider it acceptable to lose control of AI?

No. He distinguishes between misuse and known risks from catastrophic threats, such as a serious loss of control over the systems.

Is he opposed to any regulation?

No. Altman advocates for rules focused on the most advanced models and cites testing, transparency about incidents, and accountability mechanisms as essential elements.