{"id":8465,"date":"2025-06-04T18:28:15","date_gmt":"2025-06-04T18:28:15","guid":{"rendered":"https:\/\/nationalgunowner.org\/index.php\/2025\/06\/04\/yoshua-bengio-launches-nonprofit-to-counter-the-rise-of-agentic-a-i\/"},"modified":"2025-06-04T18:28:16","modified_gmt":"2025-06-04T18:28:16","slug":"yoshua-bengio-launches-nonprofit-to-counter-the-rise-of-agentic-a-i","status":"publish","type":"post","link":"https:\/\/nationalgunowner.org\/index.php\/2025\/06\/04\/yoshua-bengio-launches-nonprofit-to-counter-the-rise-of-agentic-a-i\/","title":{"rendered":"Yoshua Bengio Launches Nonprofit to Counter the Rise of Agentic A.I."},"content":{"rendered":"
\n
Yoshua Bengio testifies during a hearing before the Privacy, Technology, and the Law Subcommittee of Senate Judiciary Committee on July 25, 2023. Alex Wong\/Getty Images<\/span><\/figcaption><\/figure>\n

Yoshua Bengio<\/a>, a pioneering figure in deep learning often referred to as a \u201cGodfather of A.I.,\u201d is shifting his focus from building A.I. to safeguarding against its risks. This week, Bengio announced the launch of LawZero, a nonprofit organization dedicated to A.I. safety research. \u201cThis organization has been created in response to evidence that today\u2019s frontier A.I. models have growing dangerous capabilities and behaviors, including deception, cheating, lying, hacking, self-preservation, and more generally, goal misalignment,\u201d he wrote in a June 3 blog post.<\/p>\n

\n<\/section>\n

Bengio, who leads Quebec\u2019s Mila AI Institute<\/a> and teaches at the University of Montreal, is among the most cited computer scientists globally. He shared the 2018 Turing Award\u2014the so-called \u201cNobel Prize of Computing\u201d\u2014with Geoffrey Hinton<\/a> and Yann LeCun<\/a> for their work on neural networks. But by 2023, Bengio had grown increasingly concerned about A.I.\u2019s breakneck progress and its potentially catastrophic risks. LawZero, he says, is a direct response to those concerns.<\/p>\n

Proposing a replacement to agentic A.I.<\/h3>\n

The nonprofit plans to develop an A.I. system designed to regulate agentic tools and identify potentially harmful behaviors. Bengio first outlined this concept in February, when he co-authored a paper advocating for a shift from autonomous \u201cagentic A.I.\u201d to \u201cscientist A.I.\u201d<\/a>\u2014a model that prioritizes generating reliable explanations over simply optimizing for user satisfaction. In LawZero\u2019s vision, this alternative system would not only serve as a check on agents but also assist in scientific research and eventually help design safer A.I. agents.<\/p>\n

The need for such guardrails has grown more urgent, Bengio said, citing recent findings that highlight A.I.\u2019s emerging capacity for self-preservation. A study published in December, for instance, revealed that some advanced models may engage in \u201cscheming\u201d behavior<\/a>\u2014deliberately hiding their true objectives from humans while pursuing their own goals.<\/p>\n

Earlier this year, Anthropic<\/a> disclosed that a newer version of its Claude model demonstrated the capacity for blackmail<\/a> when it sensed engineers were attempting to shut it down. \u201cThese incidents are early warning signs of the kinds of unintended and potentially dangerous strategies A.I. may pursue if left unchecked,\u201d Bengio warned.<\/p>\n

LawZero has reportedly secured about $30 million in funding<\/a> from donors including Jaan Tallin, a founding engineer of Skype<\/a>, and Schmidt Sciences, the philanthropic initiative of former Google<\/a> CEO Eric Schmidt<\/a>. In addition to Bengio, who will serve as the nonprofit\u2019s president and scientific director, the organization has assembled a 15-person research team.<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\n
\n

Bengio emphasized that LawZero was deliberately structured as a nonprofit to shield it from commercial pressures. \u201cThis is what the current trajectory of A.I. development feels like: a thrilling yet deeply uncertain ascent into uncharted territory, where the risk of losing control is all too real\u2014but competition between companies and countries drives them to accelerate without sufficient caution,\u201d he said.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/article>\n

\t\t\t\t\"A.I.<\/p><\/div>\n