On April 7, 2026, Anthropic announced Claude Mythos Preview — a model the company describes as its most capable yet, with agentic coding and reasoning skills that exceed any prior Claude release. Anthropic judged the model too dangerous to release publicly. Instead, it launched Project Glasswing, a closed initiative giving roughly 50 organizations — including Microsoft, Apple, Amazon, Cisco, and Nvidia — access to the model specifically to find and fix vulnerabilities in critical software infrastructure.
The announcement was not targeted at enterprise workflow software. It did not mention ServiceNow, Salesforce, or Atlassian. It did not have to.
The Mechanism
The selloff in NOW, CRM, and TEAM is not a reaction to any specific business news from those companies. It is a reaction to what Mythos represents about the trajectory of AI agents in the enterprise.
Mythos Preview was not fine-tuned for cybersecurity. Its offensive and defensive security capabilities emerged from general-purpose improvements in agentic reasoning and code execution. That is the market’s problem with it. A model that can autonomously scan codebases, identify multi-step vulnerabilities, and write working exploits — without being trained to do so — signals a level of agentic competence that extends well beyond the security domain.
The implication investors have been pricing all year is blunt: if AI agents can execute complex, multi-step technical workflows autonomously, the addressable market for per-seat enterprise software contracts compresses. ServiceNow sells workflow orchestration. Salesforce sells CRM and data management. Atlassian sells developer collaboration. All three businesses depend, at their core, on human operators using software to do things. Every credible demonstration that AI can do those things without a human in the loop — or with dramatically fewer humans in the loop — reprices that dependency.
The Context
The Mythos announcement arrived into a sector already under sustained pressure. What Wall Street has been calling “SaaSpocalypse” began in earnest in early 2026, accelerated by a leaked Fortune 50 internal memo outlining plans to cut Salesforce and ServiceNow license spend by 60% through replacement with AI API credits.
ServiceNow had already dropped 10% on April 2 — days before the Mythos announcement — on a combination of federal budget pressure from DOGE-driven contract cuts and a Stifel Nicolaus price target reduction. Salesforce year-to-date losses heading into April were approaching 35%. Atlassian had shed half its value from peak. Against that backdrop, Mythos did not create the sell-off narrative. It confirmed it.
The market has now seen multiple consecutive Anthropic releases — Claude Cowork, Claude Code Security, Managed Agents, and now Mythos — that each widen the set of enterprise workflows AI can execute without traditional software mediation. Each release produces a fresh leg down in SaaS multiples. Thursday’s session was the same trade.
The Bear Case in Brief
The bear case for NOW, CRM, and TEAM is structural, not cyclical. It runs as follows: enterprise software pricing is seat-based because knowledge work is seat-based. AI agents do not occupy seats. An organization that deploys AI agents to handle IT service management, customer relationship workflows, or developer task coordination does not need the same number of seats — or, in limit cases, any seats at all. If incumbents cannot reprice their products around AI-native consumption models fast enough, their revenue growth slows, their multiples compress, and their ability to fund competitive AI development deteriorates. The slide becomes self-reinforcing.
The Bull Case
The counterargument, represented most visibly by Wedbush analyst Dan Ives, holds that the fear is mispriced and the sell-off is overdone. After CIO-level checks across the industry, Wedbush maintains that switching costs, data lock-in, and mission-critical integration give platforms like Salesforce and ServiceNow durable moats. AI agents need clean, unified data to function without hallucinating. ServiceNow and Salesforce are precisely the systems of record that provide that foundation. An enterprise that deploys AI agents without governance rails is accumulating liability, not efficiency.
There is also a monetization argument. ServiceNow is already evolving from an IT ticketing platform into an AI orchestration layer. Salesforce has repositioned around its Data 360 and Agentforce products. If AI increases the value delivered per seat rather than eliminating seats, the revenue story changes from erosion to expansion. ServiceNow’s own deepening partnership with Anthropic — embedding Claude directly into its AI Platform — suggests the company is betting on that outcome.
Assessment
The small declines in NOW and CRM on Thursday — roughly 1.4% and 0.3% respectively — are modest against the backdrop of the broader 2026 selloff in the sector. They are reflex trades on a headline, not a repricing event. But they represent something the market has been doing consistently every time Anthropic makes an announcement: treating each capability expansion as confirmation that the SaaS disruption thesis has a pulse.
Whether that thesis resolves as an extinction event or a transformation opportunity depends on a question none of yesterday’s sellers can answer: whether these platforms can shift their monetization models to match AI-native consumption patterns before their competitors do it for them.
Mythos Preview did not answer that question. It just made it more urgent.