Live·Open questions in longevity research
All news
EcosystemIndustry & policy

Raymond Douglas proposes funding independent critics of AI developers

21 November 2026· 261121006

Raymond Douglas proposes funding independent critics of AI developers

In an essay published on 16 November, the AI safety researcher at ACS Research describes an imbalance: collaboration with developers attracts more resources than public pressure on them. He suggests that those hoping to change companies from within should specify in advance what would lead them to reconsider that strategy.

One way to influence developers is through collaboration: gaining access to models, understanding a company's decisions, and proposing changes. Another is to build political support and demand restrictions. Douglas supports combining both approaches. He himself coauthored a study published by AI developer Anthropic on how interactions with AI can change users' beliefs and decisions.

In his account, collaboration attracts people by offering computing resources, information, high salaries, and capable colleagues. An individual's contribution to reducing the risk of an AI catastrophe is difficult to assess, so people begin to judge competence by visible indicators: publications and employment at a prominent company. Those who sincerely favor collaboration receive more resources and attract others to the same approach.

Public criticism, Douglas argues, can damage relationships with influential people. If criticism changes a company's direction, others share the benefits, while the critic bears the risk. A cautious researcher has more to gain by building influence and voicing concerns privately. When many people make this choice, fewer advocate public pressure than support it in private conversations.

Douglas argues that grantmaking reinforces this caution: when deciding whether to award a politically controversial grant, a foundation employee worries about their own reputation and access to funding. Cautious funding decisions leave public critics with fewer resources. On 18 November, Dan Hendrycks of the Center for AI Safety reported that many AI safety groups had privately agreed with a statement criticizing Anthropic but had declined to sign it because of the risks to their relationships and future funding.

In the discussion of the essay, AI safety researcher Zephaniah Rouw added an observation: newcomers often sincerely want to work at Anthropic. On their choice of safety strategy, he writes: “Most often, people simply haven't formed their own views on this question yet.” Douglas acknowledged that the people he had spoken with were an unrepresentative group.

Douglas proposes giving several respected researchers with unconventional views five years of funding so that public criticism poses less of a threat to their work. Greater corporate transparency would allow critics to obtain information without working closely with developers. Public critics have their own rewards, Douglas notes: media attention and audience approval, which they may begin to cater to.

He suggests that advocates of collaboration specify in advance the beliefs underlying their approach and the events that would lead them to reconsider it. If someone had planned to shift to public pressure later but subsequently decided that doing so would be pointless, Douglas asks them to announce that decision publicly. Others could then take that change of position into account when planning their own actions.

Originally published on Telegram by Ukhvat NewsView on Telegram
Sources
#ai-safety#independent-critics#research-funding#public-pressure#corporate-transparency