Meta's most-popular former employee and father of AI Yann LeCun calls Anthropic's latest model that has everyone scared, as 'Drama'

Anthropic's Claude Mythos Preview is being touted as capable of finding thousands of high-severity vulnerabilities, but Yann LeCun, Gary Marcus, and other critics say the reaction is overstated and the model is only incrementally better than predecessors. Cybersecurity firms Cisco, CrowdStrike, and Palo Alto Networks reportedly see it as a genuine inflection point, while Anthropic's revenue is said to have tripled to over $30 billion and both Anthropic and OpenAI are reportedly exploring IPOs. The debate centers on whether the product is a real cybersecurity breakthrough or a restricted, high-priced sales pitch.

Analysis

The market is discounting the headline as a binary AI-safety debate, but the more important effect is distribution: restricted access to a capability that can materially compress vulnerability discovery cycles should disproportionately benefit security vendors with the deepest enterprise workflows and incident-response footprints. If AI meaningfully shortens exploit development and patch validation times, the budget battle shifts away from point tools toward platforms that can ingest, triage, and remediate at machine speed. That is structurally positive for CSCO, CRWD, and PANW, while also strengthening the case for AI-integrated security bundles at the hyperscalers.

The second-order risk is that the “dangerous to release” framing becomes a pricing and governance moat for Anthropic-like model providers, but only for a narrow window. Over 3–6 months, the bigger trade is not model superiority; it is whether CISOs and boards force accelerated security spend after internal red-teaming demos. If that happens, the revenue impulse should show up first in consumption-sensitive security names, then in longer-cycle enterprise platform refreshes.

Meta is the odd loser in the set: LeCun’s skepticism may read as reputational damage for the broader AI-safety narrative, but it does not create a direct operating benefit. More importantly, if the market decides this is mostly marketing, the multiple expansion embedded in restricted model access could compress for private AI labs and late-stage IPO candidates, while large strategics with access get a modest negotiating advantage on procurement. The consensus may be underestimating how quickly this becomes a budget reallocation story rather than a pure sentiment story.

Key Decisions for Investors

• Go long CRWD on a 1-3 month horizon; pair against a short basket of lower-quality software names with weak security exposure. Risk/reward favors upside if enterprise breach-prevention budgets reaccelerate after board-level AI risk briefings.
• Initiate a tactical long CSCO / short META pair for 4-8 weeks. CSCO has direct monetization if AI-assisted security workflows move from pilot to deployment; META is more exposed to perception-driven skepticism with limited near-term earnings linkage.
• Add to PANW on weakness, targeting 3-6 month upside. Use a call spread rather than outright equity if you want defined risk: the thesis is not immediate monetization, but higher renewal and platform attach rates as AI threat-modeling becomes table stakes.
• Buy AAPL and GOOGL modestly as beneficiaries of restricted enterprise access and trusted-distribution status, but size smaller than cybersecurity longs. Upside is slower and mostly strategic; downside is if the narrative collapses into hype and the premium for access evaporates.
• For more convex exposure, consider a 2-4 month call spread in CRWD or PANW rather than chasing the news. The key risk is rapid mean reversion if external validation shows the capability is replicable by cheaper models, which would cut the urgency trade before budgets are approved.