As anxiety mounts over artificial intelligence safety, a number of leaders who helm frontier labs have made clear that they want — and need — better industry standards to contain the potentially dangerous technology.
Yet, like any industry leader hoping to achieve the most favorable conditions for their companies, AI executives like Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Google DeepMind co-founder Demis Hassabis are hoping to play a significant role in determining the standards they seek, say experts.
“I think what has happened recently is that they’ve fully come to terms with the fact that they actually can’t keep these things safe,” Anthony Aguirre, CEO and president of the Future of Life Institute told Mashable. The nonprofit organization focuses on preventing extreme large-scale risks of transformative technologies.
Yet, Aguirre added, the fresh candor comes with a catch: “Certainly they want to be in control of the process.”
What that could look like for leading AI companies is self-regulation, along with transparency and accountability on their terms, plus assurances about antitrust law.
AI companies want artificial general intelligence
Aguirre said that despite recognizing perilous emergent safety concerns, major AI companies still won’t abandon their goal of achieving artificial general intelligence, or AGI.
This form of AI would match or exceed human intelligence, and it could lead to the replacement of many human jobs with automated ones along with the rise of AI decision makers overtaking human leadership, Aguirre explained.
“They want to sort of have a machine-run future,” Aguirre said. “This is not something most people want.”
Calls for pacing the frontier, as Amodei recently described in a personal essay on AI safety risks, don’t involve pausing the pursuit of AGI.
Aguirre believes that’s a mistake. Instead, he said AI companies should “totally change direction” away from trying to create AGI, though he realizes they’ll likely refuse to do so.
What the AI companies want in exchange for safety standards
Self-regulation
Pete Furlong, senior manager of policy and research at the Center for Humane Technology, told Mashable that the industry has been pushing for self-regulation for some time.
“We’ve seen them talk about the need for regulation in certain aspects, but I don’t think their actions always necessarily back that up,” he told Mashable. Furlong added that the top companies want to protect their “industry leader status” and may seek ways to operate in a “closed door system” without much public input on risks and solutions.
In a July Substack post on safety, Google’s Hassabis suggested the U.S. create a standards body modeled on existing federally overseen public-private partnerships or self-regulatory organizations. Using the Financial Industry Regulatory Authority as an example, Hassabis said the standards body would lead on developing model assessment protocols and collaborating with federal agencies specifically on national security testing.
In his essay, Amodei wrote that Anthropic has supported “sensible and targeted AI regulation,” particularly on transparency and third-party auditing. Acknowledging the lengthy process of passing laws, Amodei also backed a voluntary effort on behalf of AI companies to set standards subsequently verified by “embedded” third-party evaluators. He unilaterally committed Anthropic to relying on those evaluators in its safety work.
Still, experts cautioned against accepting this safety measure at face value.
“While AI developers’ support for independent evaluation is a real step forward, additional mechanisms will be required for effective accountability,” Matt O’Shaughnessy, senior technologist at the Center for Democracy and Technology’s AI Governance Lab, told Mashable. He added that it would be a “net-negative for accountability” if AI oversight amounted only to independent assessments.
Mashable Light Speed
Limited transparency and accountability
Furlong credited Anthropic for recently releasing a comprehensive safety report on AI-enabled risks like bioweapons and mass surveillance. He also said it was “laudable” that OpenAI had independent researchers investigate the Hugging Face incident during which an OpenAI agent hacked another company. Yet, OpenAI also placed “significant restraints” on that investigation, including time-frame limits, Furlong said.
While these are reasonable steps toward transparency, Furlong believes it’s fair to expect more.
“It can’t be entirely up to these companies to dictate the terms of their transparency,” he said.
While OpenAI and Anthropic have both backed transparency in writing, Aguirre said AI companies’ actions don’t always align with their public stance.
“I think there is a big gap between what they’re saying and what their lobbying groups behind closed doors do,” Aguirre said.
The major AI companies have also fought attempts to hold them accountable for worst-case scenarios, like when a user consults their chatbot about committing a mass shooting, gets life-threatening medical advice from it, or confides in it about feeling suicidal and receives advice on how to take their own life.
“There’s a lot of potential concerns around companies trying to skirt any sort of meaningful accountability,” Furlong said.
Reassurance on antitrust law
As the leading AI companies convene to discuss safety problems and solutions, they also express worry that doing so could expose them to anti-trust lawsuits.
Amodei said in his recent essay that the government should mediate or enable standard-setting discussions. He added that the government needs to issue a “narrow waiver” permitting specific safety conversations.
Aguirre said he agrees with the companies in this instance. Provided the waiver is carefully written to apply to safety discussions, and not coordinated activity that would freeze out smaller AI companies, Aguirre believes it’s reasonable to want clarity about antitrust laws.
Indeed, a class-action lawsuit filed this week by Trial Lawyers for Justice alleges that a coordinated slowdown by Anthropic, OpenAI, Google, and SpaceXAI violated antitrust law.
The main takeaway
Aguirre believes that while many AI leaders are legitimately worried about safety, their proposals don’t offer a radically different vision of what comes next.
“These companies are building something that they know that they cannot keep safe or under control, and they’re doing so without any of the normal oversight or regulation that we apply to every other potentially unsafe industry,” Aguirre said. “They’re asking us to let them do the same thing, but just help them out a little bit to do it a little bit more safely.”
Additional reporting by Chase DiBenedetto.
Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
