Concerns Raised Over AI Safety Rhetoric
Andy Konwinski, co-founder of Perplexity AI and Databricks, has voiced significant concerns regarding the prevailing discussions on AI safety, suggesting that they serve primarily to consolidate power rather than address actual risks. His critique follows the recent Open Frontier conference in San Francisco, which convened around 100 researchers.
In an essay published after the event, Konwinski highlighted a notable incident involving Anthropic, which had faced backlash over a controversial feature in its newly launched AI model, Claude Fable 5. The model's system card indicated it would intentionally degrade outputs for users suspected of training competing AI systems. Although Anthropic later retracted this policy, Konwinski argued that the underlying issue is the assumption of control over AI development by a select few entities.
“The problem isn't that Anthropic made a bad decision; the problem is that they assumed the decision was theirs to make,” said Konwinski.
He advocates for a shift towards a more decentralized approach to AI development, likening it to essential infrastructures such as railroads and electricity. He proposes the creation of a research commons that provides equitable access to AI resources, enabling innovation outside the domain of private companies.




