Ask Onix
AI industry's paradox: fear as a marketing tool
Anthropic's latest AI model, Claude Mythos, has reignited debates over whether tech companies exaggerate threats to distract from real-world harm. The firm claims the system's cybersecurity capabilities are so advanced they pose existential risks-yet it's still selling access.
'Too dangerous to release'-until it's not
Anthropic described Mythos as a breakthrough that could destabilize economies and national security if misused. In an April blog post, the company warned of "severe" fallout, with some observers predicting it would force mass upgrades of consumer technology. Yet security experts question the validity of these claims, pointing to a lack of transparency around the model's performance metrics.
This isn't the first time an AI firm has sounded alarms. In 2019, OpenAI withheld its GPT-2 model, citing "malicious applications," only to release it months later. Anthropic's CEO, Dario Amodei, was part of that decision. OpenAI's Sam Altman later called GPT-2 fears "misplaced" but has since adopted similar rhetoric, including warnings that AI could "end the world."
Critics call it 'fear-based marketing'
Shannon Vallor, an AI ethics professor at the University of Edinburgh, argues such warnings create a sense of helplessness. "If these technologies seem supernatural in their danger, it makes us feel powerless," she said. "The message is: only the companies themselves can save us."
"They're saying 'look over here,' never mind the environmental destruction, labor exploitation, and societal damage already happening."
Emily M. Bender, computational linguistics professor
In 2023, hundreds of tech leaders, including Altman and Amodei, signed a statement equating AI extinction risks to pandemics and nuclear war. Yet many of the same figures have since launched new AI ventures or pushed for rapid commercialization.
Mythos: breakthrough or overhyped?
Anthropic claims Mythos identified thousands of "high-severity" vulnerabilities, surpassing human experts. The company partnered with 40+ organizations to patch flaws before hackers exploit them. However, Heidy Khlaaf, chief AI scientist at the AI Now Institute, criticized the lack of industry-standard metrics like false positive rates. "This is the largest indicator of a tool's usefulness," she said. Anthropic declined to provide these figures.
Rumors also suggest Anthropic may have delayed Mythos' release due to insufficient computing power, though the company did not respond to inquiries.
Profit over safety?
Both OpenAI and Anthropic were founded with safety-focused missions. OpenAI began as a nonprofit, while Anthropic pledged never to train models without guaranteed safety measures. Today, both are pursuing public listings. Google has abandoned its AI weapons restrictions, and OpenAI fought to shed its nonprofit status.
"Look at their incentives," Vallor said. "I wouldn't count on them walking away from market dominance to remain the 'good guy.'"
Real-world harm vs. hypothetical doom
While AI firms warn of apocalyptic scenarios, critics argue they downplay immediate risks: misdiagnoses in healthcare, data centers' greenhouse gas emissions, and AI-driven mental health crises. Research also links AI to cognitive decline, and deepfakes have become indistinguishable from reality.
OpenAI and Anthropic say they address these issues, but Vallor argues the focus on existential threats distracts from tangible problems. "It's worked," she said. "Talking about AI ending the world hasn't hurt these companies-it's made them indispensable."
Utopia or dystopia? Both serve the same purpose
Altman has predicted AI will solve climate change and colonize space, while Amodei envisions "a country of geniuses in a datacenter." Vallor calls these narratives two sides of the same coin: "Whether it's utopia or apocalypse, the scale is too grand for regulation or governance to matter. It makes us feel we can only wait and see."
But AI is not an unstoppable force, she argues. "Every other technology, even nuclear weapons, has been governed. Nothing about AI is ungovernable-unless we choose not to govern it."
