OpenAI's Altman says AI benefits justify some risk, rejects catastrophic outcomes
In short
OpenAI CEO Sam Altman said in a Decoded interview that AI's benefits are worth accepting some risk and that the technology should stay widely open to the public, while adding that he does not accept truly catastrophic risk such as loss of human control over AI. He described a “huge gulf” between OpenAI and Anthropic, whose CEO Dario Amodei has pushed strict AI regulation, though the documents note the two companies' positions on external safety evaluation have recently drawn closer.
Read the full story 2 min read
OpenAI chief executive Sam Altman said the benefits of artificial intelligence are worth accepting some risk and that the technology should remain widely open to the public, in an interview with the Decoded programme. “We believe that in order to get the benefits of this technology and give people autonomy, the world should accept certain costs and negative events that come with it,” he said, according to IT Home and 36Kr. CLS reported that Altman disagreed with rival Anthropic on AI risk, saying he does not accept that AI poses an “extremely destructive risk”. [ 1 , 3 , 4 ]
Asked where he differed from Anthropic and its chief executive Dario Amodei, Altman said “I think there is a huge gulf between the two sides”, IT Home reported. IT Home noted that Anthropic has long been seen as the industry's most committed advocate of strict AI regulation, while OpenAI previously favoured lighter rules. [ 3 ]
Altman also drew a boundary around the risk he accepts. He said AI's eventual positive outcomes should exceed negative effects by “several orders of magnitude”, but that he does not accept genuinely catastrophic risk, including severe loss of human control over AI. According to cnBeta, the interviewer described the distinction as key: the “bad things” Altman accepts mainly refers to individual serious incidents that may accompany technology diffusion and deployment, not allowing systems to cause catastrophic consequences. [ 2 , 3 ]
The documents say the two companies' positions are not fixed. cnBeta reported that Altman publicly endorsed Amodei's call to control the pace of frontier model development last month and supported giving independent evaluators employee-level access, and that OpenAI has backed a bipartisan US congressional proposal requiring top AI companies above certain revenue and compute thresholds to undergo external safety assessments. IT Home said OpenAI recently moved toward supporting stricter industry rules as it dealt with high-profile hacking attacks and model misbehaviour. [ 2 , 3 ]
On state-level regulation, cnBeta reported that OpenAI previously pushed for uniform federal rules and opposed a patchwork of state laws, but that in September it announced support for four California AI safety bills covering independent risk assessment bodies, standards for AI auditors, protection for teenagers using chatbots and defences against AI biological threats. The company also said it supports national safety requirements graded by model capability and risk, and that states can keep moving before federal legislation is passed, cnBeta reported. [ 2 ]
Taken together, cnBeta and IT Home said OpenAI and Anthropic still differ on how strict regulation should be and how fast development should proceed, but that their distance on external evaluation and controlling frontier risk is narrowing. [ 2 , 3 ]
Why it matters
The remarks put on record the differences between two of the most prominent AI developers over how much risk society should accept as the technology advances — a debate that shapes safety rules in the United States and elsewhere. The documents also indicate the gap between them is narrowing on external evaluation of frontier models, even as they still differ on how strict regulation should be and how fast development should proceed.
Key facts
- OpenAI CEO Sam Altman said the benefits of AI are worth accepting some risk and that the technology should remain widely open to the public. [ 3 , 4 ]
- Altman said he disagrees with Anthropic on AI risk and described a “huge gulf” between the two companies' worldviews on AI regulation. [ 1 , 3 ]
- Altman said he does not accept genuinely catastrophic risk, including severe loss of human control over AI. [ 2 , 3 ]
- Altman said AI's eventual positive outcomes should exceed negative effects by “several orders of magnitude”. [ 2 ]
- cnBeta reported that Altman last month publicly endorsed Anthropic CEO Dario Amodei's call to control the pace of frontier model development and supported giving independent evaluators employee-level access. [ 2 ]
- Documents state that OpenAI and Anthropic's regulatory positions have been converging in recent months, though they remain divided on regulatory strictness and development speed. [ 2 , 3 ]
- cnBeta reported that in September OpenAI announced support for four California AI safety bills covering independent risk assessment bodies, AI auditor standards, protection for teenagers using chatbots and defences against AI biological threats. [ 2 ]
Confirmed by several sources
- Altman said AI's benefits justify accepting some risk and that the technology should remain widely open to the public. [ 3 , 4 ]
- Altman said he does not accept catastrophic risk, including severe loss of human control over AI. [ 2 , 3 ]
- Altman's remarks were made in an interview in which he was asked about his differences with Anthropic and its CEO Dario Amodei. [ 3 , 4 ]
- Documents state that OpenAI and Anthropic positions have been converging in recent months even as disagreements remain. [ 2 , 3 ]
Still unclear
- The specific regulatory measures OpenAI is said to support The four California AI safety bills, the bipartisan congressional proposal requiring external safety assessments for top AI companies above revenue and compute thresholds, and continued state rulemaking before federal legislation are reported by cnBeta alone and are not confirmed by other documents.
- The exact scope of the risk Altman says society should accept Altman said the world should accept certain negative events; cnBeta's account explains this refers mainly to individual serious incidents accompanying technology diffusion and deployment rather than catastrophic outcomes, but the documents give no further detail on which incidents he would consider acceptable.
What local media are saying
Timeline, local time
- CLS reports that Altman's position on AI risk diverges from Anthropic's and that he does not accept an “extremely destructive risk”. [ 1 ]
- cnBeta publishes a fuller account of the interview, including Altman's risk boundary and OpenAI's regulatory positions. [ 2 ]
- IT Home publishes Altman's remarks, including the “huge gulf” comment and his view that AI's benefits are worth accepting some risk. [ 3 ]
- 36Kr reports Altman's remarks that AI's benefits justify accepting risk and that the technology should remain broadly open. [ 4 ]