When Yoshua Bengio speaks about AI risk, the industry listens. The Turing Award winner and one of the three godfathers of deep learning sat down with Bloomberg Businessweek Weekend on June 5, 2026 and told viewers the field has crossed a threshold it cannot yet reason about.
His exact words: “I think I should have been seeing ahead, you know, earlier that we were building something that could become extremely powerful and that we don’t know how to control.”
The admission lands as autonomous AI agents already operate independently toward their own goals, and as Anthropic’s Mythos model has triggered emergency responses from intelligence agencies for its demonstrated cyber capabilities.
Why The Quote Matters For Investors
Bengio framed the commercial pressure clearly. Unlike nuclear weapons, AI development carries “massive financial incentives pushing companies towards going faster and faster and not paying enough attention to the safety issues.”
That dynamic sits at the center of the AI investment thesis. The same capital fueling revenue at NVIDIA (NASDAQ: NVDA | NVDA Price Prediction) and Amazon (NASDAQ: AMZN), which holds a strategic stake in Anthropic, also pressures labs to compress safety timelines. Bengio noted that even if one firm develops safeguards, competitors could reach similar capability within 6 to 12 months.
That window has direct portfolio implications. Hyperscaler capex, GPU demand, and inference revenue all depend on continued model release cadence. Any regulatory pause, whether voluntary or imposed, would compress those forecasts.
The Specific Risks He Named
Bengio enumerated capabilities companies are tracking internally: sophisticated cyberattacks, biological engineering that could engineer dangerous viruses and create new pandemics, and AI building the next generation of AI while potentially inserting backdoors to create systems “not friendly” to humans.
He pointed to Anthropic’s Mythos as the working example. UK AI Safety Institute evaluations of Claude Mythos Preview confirmed continued improvement in capture-the-flag style hacking challenges, and Bain has publicly described the model as a cybersecurity wake-up call. Mythos itself is private, though the public AI supply chain (compute, networking, energy) reflects every dollar Anthropic and its peers spend chasing the frontier.
The dual-use point he returned to: the same systems enabling breakthroughs in medicine, batteries, and agriculture also enable dictatorships, terrorist attacks, and infrastructure collapse. Anyone evaluating AI exposure has to price both sides.
What He Is Building
In June 2025, Bengio launched LawZero, a non-profit AI safety lab designed to develop safe-by-design AI systems and to prioritize safety over commercial imperatives. He has described today’s trajectory as driving up an unfamiliar mountain road in thick fog with no guardrails.
That framing explains his answer when Bloomberg pressed him on whether the positives outweigh the negatives: “Well, there’s so much we don’t know here. You have to weigh in the uncertainty.”
What To Watch Next
Bengio called for international coordination, saying “initially, it’s probably going to be just the US and China.” That puts three near-term catalysts on investor radar:
- US export control updates targeting frontier model training compute
- China’s response to any joint safety framework, particularly around its own frontier labs
- Industry-wide pause proposals, including Anthropic’s public call for a mechanism to slow development
Any of those moves would reset the risk premium on the AI capex cycle. The names most exposed are the picks-and-shovels providers and the hyperscalers funding Anthropic, OpenAI, and Google DeepMind. NVIDIA’s data center revenue trajectory remains the cleanest single read on whether the industry is, in fact, slowing down.
Bengio’s quote reframes the question every AI investor is implicitly answering, even as the build-out continues: how much of today’s valuation assumes continued, unconstrained scaling, and how much room is there if Washington, Beijing, or the labs themselves decide that pace needs to change.
Top Gaining Stocks
Top Losing Stocks