Elon Musk: Slowing Down Safety and Conspiracy Lawsuits Leave AI Under Fire
Elon Musk responded to David Sacks' lengthy post about "Trust and Safety 2.0," stating that while some claim major labs are not slowing down enough, others are suing these companies for openly discussing slowing down for safety, facing criticism whether they act or not.
Sacks described the new pressure as embedded effective altruism oversight that is difficult to dismiss in every AI company. He compared it to the previous round: after media-induced panic, social platforms formed trust and safety teams, collaborating with organizations like the Southern Poverty Law Center to review conservative and dissenting voices, bypassing the First Amendment; until Musk acquired Twitter and publicly released Twitter files, this was not revealed. He claimed this review would be more comprehensive and covert, and if demands are not met, they would call the media to say "the company is risking humanity," and be celebrated as whistleblowers. The post pointed to aievaluatorforum.org.
Musk's addition framed the conflict as a hedge between legal and public opinion: slowing down could be framed as inaction, while publicly coordinating to slow down could be framed as antitrust conspiracy. He did not name specific plaintiffs or case numbers, only pointing out the existence of two opposing accountability frameworks on the same issue.
Sacks is currently the co-chair of the President's Council of Advisors on Science and Technology and previously served as the AI and Crypto Czar. In a recent interview, he placed safety responsibilities primarily on developers like Anthropic and OpenAI, stating that if they cannot control it, they should not do it, and they should not rely on creating public fear to gain regulation. There are internal divisions in the White House regarding the tightness of AI regulation, with Sacks and some Silicon Valley executives favoring a lighter touch, while other officials advocate for more scrutiny.
This set of tensions arises as product liability, safety assessment outsourcing, and media whistleblower narratives heat up, not as a new legislative text.
In market mechanisms, this marks the "personnel authority of the safety layer" as a political asset. What is sold is external oversight and the narrative of slowing down, while what is bought is the labs setting their own pace with less mandatory evaluations from the government. Funding and computing power still follow the release calendar; the beneficiaries are labs and investors opposing making effective altruism assessments a permanent veto, while the pressured parties are evaluation organizations that rely on safety crises for staffing and media positions.
Being attacked from both sides does not change whether model training continues, only who has the authority to define "safe enough."
Source: Public Information
ABAB AI Insight
Musk used a proverb to overlay two years of lawsuits and safety lobbying on the same picture: xAI needs to catch up, OpenAI has been sued by him for slowing down competitors; meanwhile, the safety faction wants labs to publicly commit to slowing down. Both sides need him to take a stand, and he chose to point out the incompatible incentives. Sacks provided an organizational history: trust and safety have shifted from social content moderation to model release gates, with personnel changing from platform employees to external overseers in labs, and accountability shifting from the board to media headlines.
The capital path is that evaluation organizations want permanent staffing and "cannot be fired," while labs want release rights. Whoever controls access controls the release schedule. The Twitter files were treated as evidence from the previous round: after the review layer merged with external organizations, the First Amendment does not cover private enterprise policies. This time, the battlefield has shifted to model weights and system cards.
A similar structure exists with drug safety officers embedded in pharmaceutical companies, and content moderation outsourced to fact-checking networks. AI governance is in a phase of power struggle: technology is still expanding, and veto power is competing for positions.
This belongs to the personnel battle before regulatory changes. The mechanism is that "expert attire" writes policy preferences as safety science, then uses whistleblower honors to lock in media; on the other side, the same group of people is portrayed as a review layer bypassing elections. Being attacked from both sides is because safety and competition policies cannot be satisfied with the same set of incentives simultaneously.
ABAB News · Cognitive Law
Slowing down is moral, coordinating to slow down is litigation
The review layer changes its name, but the question remains who can press the release button
When whistleblowers are celebrated, first check if they can be fired