Anthropic Accused of Using “Fear” to Sell Claude Mythos, Says Sam Altman
OpenAI CEO Sam Altman has dismissed growing anxiety around Anthropic’s latest AI system, Claude Mythos, arguing that the rival lab is leaning on alarmist messaging to bolster its brand and consolidate influence over advanced AI.
In a recent appearance on the Core Memory podcast with technology journalist Ashlee Vance, Altman suggested that some of the discourse around Mythos’ capabilities-particularly in cybersecurity and potential misuse-has been framed in a way that stokes fear more than it reflects the full reality.
According to Altman, this style of communication risks turning AI safety into a marketing strategy rather than a shared, evidence-driven concern.
He described the approach as a kind of “fear-based” positioning that, in his view, can be used to justify why only a narrow set of actors should be trusted to build and deploy cutting-edge AI.
Altman acknowledged that powerful AI systems do introduce real risks and that some level of caution is warranted. However, he argued that there is a difference between honestly communicating risk and using worst‑case scenarios as a tool to gain market and regulatory leverage.
In his words, it’s easy to construct an argument that “we need control of AI, just us, because we’re the trustworthy people” by emphasizing the most frightening possibilities. That framing, he implied, can be remarkably effective if the goal is to keep advanced AI development concentrated inside a small circle of companies and institutions.
Claude Mythos and the Cybersecurity Debate
Anthropic’s Claude Mythos has drawn particular scrutiny for its reported prowess in cybersecurity, including its ability to analyze code, discover vulnerabilities, and potentially automate parts of offensive or defensive security workflows.
Supporters of strong restrictions on such systems worry that increasingly capable models could lower the barrier to sophisticated cyberattacks, empower small groups or even individuals to cause outsized damage, and overwhelm existing defensive infrastructure.
Altman did not deny that a model like Mythos can be used in harmful ways. Instead, he pushed back on what he sees as selective emphasis: highlighting potential catastrophic misuse while downplaying the benefits of broader access, collaborative oversight, and multiple competing labs working to harden systems and improve defenses.
From his perspective, painting a rival model as uniquely dangerous can serve commercial interests as much as, or more than, public safety.
Safety Concerns vs. Market Strategy
Altman drew a line between “legitimate safety issues” and narratives that, in his view, overstate the threat to justify tighter control by a few players.
He conceded that advanced AI tools will inevitably require regulation, independent evaluation, and technical safeguards. But he warned that when fear becomes the primary lens, it can shape rules and norms in a way that entrenches incumbents.
The implication is that if policymakers and the public are convinced that only a handful of “responsible” actors can be trusted with powerful AI, those actors will gain disproportionate influence over the future of the technology. That, Altman suggested, can be as much about power and positioning as it is about ethics.
He characterized the current AI landscape as one where marketing narratives increasingly fuse with safety messaging, blurring the boundary between genuine risk communication and strategic brand building.
Concentration of Power in Advanced AI
Altman’s comments touch a broader anxiety in the tech world: who should control frontier‑level AI systems, and on what terms?
One camp argues that only a small, highly scrutinized group of organizations should develop and deploy the most advanced models, especially those with dual‑use capabilities in cybersecurity, biosecurity, or critical infrastructure. They claim that diffusion of such tools increases the chance of catastrophic misuse.
Another camp worries that concentrating AI capabilities in too few hands could create long‑term political, economic, and social imbalances. In that view, using fear to justify exclusivity can slow down open research, reduce competition, and give a small number of companies outsize sway over global information flows and industrial productivity.
Altman’s critique of how Claude Mythos is being framed positions him closer to the latter concern: he appears wary of safety narratives that double as arguments for centralization.
The Competitive Undercurrent
Altman’s remarks also arrive against the backdrop of intensifying rivalry among leading AI labs. Anthropic, OpenAI, and others are racing to release more capable models, secure partnerships, and win enterprise customers.
In such a climate, how a company talks about its own systems-and about competitors’-is not neutral. Messaging around safety, responsibility, and risk can shape:
– regulatory expectations,
– investor sentiment,
– enterprise adoption, and
– public trust.
By calling out what he describes as fear‑driven messaging around Claude Mythos, Altman is not only commenting on AI ethics but also contesting the narrative terrain on which these companies compete.
His stance suggests that safety talk cannot be fully separated from market strategy: how risk is framed can either open the field to many players or justify its restriction to a select few.
The Double-Edged Role of Fear in AI Discourse
Fear has always been a powerful force in technology debates. In AI, it appears in two forms:
1. Substantive fear – grounded in plausible scenarios of misuse, systemic disruption, and long‑term societal impact.
2. Instrumental fear – used rhetorically to push for specific outcomes, such as stricter controls, favorable regulation, or reputational advantage.
Altman’s argument implies that Anthropic’s talk around Claude Mythos leans too far into the second category, even if it is anchored in real technical concerns.
The challenge for the industry is that completely separating these two roles is nearly impossible. When companies speak about risk, they inevitably shape perceptions of themselves and their competitors.
Balancing Transparency and Hype
Anthropic has positioned itself as heavily focused on AI safety and alignment. That branding emphasizes technical safeguards, internal governance, and cautious rollout strategies. For many observers, this is a welcome counterweight to rapid, less constrained deployment.
Altman’s critique doesn’t reject the value of such caution outright; instead, it questions whether safety‑centric messaging is sometimes crafted to be emotionally charged, highlighting extreme outcomes in ways that resonate with regulators and the public-but also enhance the company’s image as the uniquely responsible steward of dangerous tools.
This dynamic leads to a paradox: the more a model is described as powerful, risky, and tightly controlled, the more it can appear both like a public hazard and a premium, elite product. Fear and prestige become intertwined.
Implications for Policy and Regulation
If policymakers internalize fear‑heavy narratives around certain models or labs, the resulting rules may:
– lock advanced development behind high regulatory barriers,
– implicitly favor well‑funded incumbents that can navigate those barriers, and
– make it hard for smaller firms or open research efforts to meaningfully participate.
Altman’s warning suggests that the way safety is communicated now could shape the structure of the AI industry for years. He appears to be advocating for a more balanced, evidence‑driven conversation that recognizes risks without turning them into a justification for permanent gatekeeping.
At the same time, his stance will be scrutinized as coming from an executive whose own company is a central player in that very power struggle.
The Broader Debate on AI Access
Underlying this dispute is a fundamental question: should the most capable AI systems be widely accessible, carefully tiered, or heavily locked down?
– Advocates of broad access argue that open or semi‑open availability fosters innovation, spreads economic benefits, and allows a wider community to discover vulnerabilities, biases, and failure modes.
– Proponents of restricted access counter that some capabilities-such as advanced cybersecurity exploitation, chemistry, or bioengineering assistance-are too dangerous to democratize, and must be kept under strict control.
Anthropic’s posture around Claude Mythos has largely aligned with the second perspective, emphasizing safeguards and controlled deployment. Altman’s comments challenge not the need for safeguards themselves, but the rhetorical move from “this is risky” to “therefore, only a tiny group should hold the keys.”
What Comes Next for Claude Mythos and Its Rivals
As Claude Mythos continues to be tested and integrated into real‑world workflows, independent evaluations of its capabilities-both beneficial and harmful-will matter more than marketing language from any lab.
Key questions for the coming months include:
– How does Mythos actually perform in cybersecurity contexts compared with other leading models?
– Do technical and policy safeguards meaningfully reduce misuse risks?
– Will regulators treat it, or similar future systems, as requiring special oversight?
– And to what extent will safety narratives continue to serve as competitive positioning tools?
Altman’s remarks ensure that, at minimum, the industry’s use of fear and risk as branding levers will receive more critical attention.
A Debate That Won’t End With One Model
The clash over how Claude Mythos is presented is unlikely to be the last argument over fear‑driven storytelling in AI. As models grow more general, more autonomous, and more deeply embedded in infrastructure, the temptation to lean on dramatic risk framing-for funding, influence, or protection-will only increase.
Altman’s intervention underscores a tension at the heart of the AI era: society needs honest, sometimes uncomfortable conversations about risk, but it also needs to recognize when those conversations are being shaped as much by competitive strategy as by concern for the public.
How that tension is resolved will influence not just the fate of Claude Mythos or OpenAI, but who ultimately gets to build, own, and govern the most powerful systems of the coming decade.