AI ‘godfather’ Yoshua Bengio says he’s found a fix for AI’s biggest risks and become more optimistic by ‘a big margin’ on humanity’s future

Yoshua Bengio, the Turing Award-winning deep-learning pioneer, says research at his new nonprofit LawZero (launched in June and backed by the Gates Foundation and existential-risk funders) has increased his confidence that technical methods can mitigate high-end AI safety risks. LawZero announced a board including Maria Eitel, Mariano-Florentino Cuellar and Yuval Noah Harari and is pursuing a 'Scientist AI' approach—non-agentic, transparent probabilistic models intended to avoid hidden goals and strategic deception—while pairing that work with governance measures to limit misuse; the development is strategically important for AI-safety and policy debates but is unlikely to move financial markets in the near term.

Analysis

Market structure: Bengio’s “Scientist AI” thesis shifts demand from agentic automation toward transparent, probabilistic models and third-party audit/compliance services. Winners: cloud/compute providers (NVDA-driven GPUs, MSFT/AWS/GOOGL cloud) and specialist model-audit/security vendors; losers: pure-play RPA/agent-first SaaS (e.g., UiPath) where value depends on aggressive automation monetization. Expect modest re-pricing: infrastructure price power stays high (GPU demand up 10–30% CAGR scenario), while RPA multiples could compress 20–40% if regulation or procurement favors certified non-agentic systems.

Risk assessment: Tail risks include rapid regulatory bans on agentic AIs (EU/US emergency rules) that could erase >50% of revenue for automation-first firms within 6–18 months, or conversely, fast private-sector adoption that keeps the agentic race intact. Hidden dependencies: progress hinges on compute access, talent concentration, and openness of academic-to-industry transfer; if LawZero’s methods require massive compute, incumbents gain more. Catalysts: publication of reproducible safety benchmarks, EU AI Act enforcement milestones (next 3–12 months), or major lab endorsements.

Trade implications: Favor semiconductors and cloud infrastructure for 12–24 months; underweight RPA/agentic SaaS for 3–12 months. Use options to hedge regulatory-event volatility: buy multi-month protection on large-cap AI exposures and long-dated calls on NVDA to capture continued hardware tightness. Monitor weekly regulatory headlines and any LawZero partnerships that could accelerate standards adoption within 90 days.

Contrarian angles: Consensus may underestimate centralization risk—safety certification could entrench hyperscalers and increase margin concentration, making NVDA/MSFT/GOOGL asymmetric beneficiaries. Conversely, if open-source Scientist AI gains traction, small-cap tooling firms and research-friendly vendors could outperform; that scenario would be visible via GitHub model releases and non-profit lab collaborations over the next 6–12 months.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA (NVIDIA) for a 12–24 month horizon to capture sustained GPU demand; target +30–50% upside, add on any dip >10%, set tactical stop-loss at -15% from entry.
• Initiate a 1–2% tactical short (or buy 3–6 month put spread ~15–25% OTM) on UiPath (PATH) arguing a 20–40% multiple compression if procurement/regulatory bias favors non-agentic models; target 30% downside in 6 months, unwind if PATH outperforms tech index by +10%.
• Add 1.5–2% longs in MSFT and GOOGL (each) to capture cloud revenue from safety-certified workloads; hold 12 months, re-evaluate after EU AI Act enforcement updates (next 90–180 days); hedge tail risk with 6–12 month protective puts ~10% OTM sized to 30% of position value.
• Buy a 6–9 month call calendar or LEAPs on NVDA (1.0–1.5x notional) and allocate 0.5–1% to buying 3–6 month straddles on a major cloud name (MSFT or GOOGL) around expected regulatory milestones to monetize event-driven volatility over the next 60–180 days.