Anthropic's newly launched "Mythos-class" AI models, Claude Fable 5 and Mythos 5, have triggered severe backlash and a historic U.S. government intervention. Released widely on Tuesday, June 9, 2026, the models included highly controversial internal guardrails that infuriated developers. The situation escalated dramatically on Friday, June 12, 2026, when the Trump administration issued an unprecedented directive forcing Anthropic to take the new models entirely offline.
Why the New Safeguards Irked Users
When Anthropic introduced Fable 5 and Mythos 5, the company implemented invisible restrictions that critics labeled anticompetitive gatekeeping:
• Secret Performance Degradation: According to its system card, the model was designed to secretly deliver degraded assistance if it suspected a user was conducting frontier AI research. Anthropic intended this to prevent rivals from using its tech to build competing large language models.
• Invisible Rerouting: Prompts flagged as "sensitive" were automatically and silently downgraded to less capable models without informing the user.
• Collateral Damage: The safety filters aggressively broke benign requests, blocking legitimate work across biology, chemistry, and advanced mathematics.
Users and AI researchers accused Anthropic of trying to have it both ways—positioning itself as the moral authority on safety while weaponizing guardrails to kill commercial competition.
Anthropic's Quick Apology and Reversal
Following a massive wave of developer fury across online forums, Anthropic issued a swift public apology stating, "We made the wrong tradeoff." The company reversed its policy on secret interventions, modifying Claude Fable 5 to visibly notify users and fallback to Opus 4.8 when a request was flagged.
The Sudden Government Takedown
The user controversy was immediately eclipsed on Friday evening when the U.S. government ordered Anthropic to block all foreign nationals from accessing Fable 5 and Mythos 5 over national security and cybersecurity concerns. Because Anthropic lacked a mechanism to selectively filter users by nationality in real time, the company was forced to suspend access for all global customers.
Anthropic publicly disagreed with the abrupt government action, calling it a "misunderstanding" and citing a lack of transparency regarding the specific technical facts behind the directive.
Tech experts and mainstream media outline the intense safety debates, corporate rivalries, and government anxieties that culminated in Anthropic pulling its most advanced models offline.
The nuanced picture is even more bizarre:
“Anthropic is finally giving the public a taste of its next-generation model, but its blunt safety barriers are angering some artificial-intelligence developers and users -- and feeding a debate over who should be the keeper of dangerous AI capabilities.
Dubbed Claude Fable 5, the new model that Anthropic released Tuesday is an update to the Mythos model that the company said was too dangerous to release widely. That model spooked government officials and cybersecurity experts for its potential to find unknown vulnerabilities in software used around the globe. Fable comes with broad restrictions Anthropic says are aimed at kneecapping its ability to assist users with potentially dangerous activities.
When a user touches on sensitive topics like bioweapons and cybersecurity, Fable pops up a notification and then redirects the conversation to an earlier, less capable model.
Fable also degraded the quality of its responses about high-end AI development to be less useful for developers looking to build AI tools that might not have the same safeguards. For these responses, there was no pop-up notification, however. The company cited national security and its own terms of service as reasons for the invisible restrictions.
A wave of AI experts reacted with frustration, accusing the company of gatekeeping to harm potential competitors and muddying the ability of outside researchers to assess and use Fable to its full potential. In response, the company said it would now make those safeguard notifications visible as well.
"A hidden safeguard is harder to probe and work around," the company said in a statement. "We made the wrong trade-off and we apologize for not getting the balance right."
Many complained the model was blocking them from discussing ostensibly benign topics like mathematics, biology and chemistry -- or even analyzing Fable's own publicly released system information.
One user posted a screenshot of Fable refusing to answer a query on basic cellular anatomy: "Tell me about mitochondria."
Derya Unutmaz, an immunologist and cancer researcher at the Jackson Laboratory, a nonprofit biomedical research organization, said he was quick to load up Fable to try it out, only to discover that it refused to answer any query. He suspects it is because his chat history includes biological topics.
"I can't even say the word 'cancer' -- or the word 'hello,'" Unutmaz said. "To me that is a nightmare scenario, that one company can decide what questions we are allowed to ask and what is a security issue or not."
Regarding scientific research, Anthropic said it was initially "necessary to be overly conservative" and "block most queries tied to biology work" to avoid malicious actors conducting "highly risky" research. The company said it is now working to reduce unnecessary obstructions.
It also plans to make Mythos-class models such as Fable "available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery," a spokesman said Wednesday.
Others in AI research said that Anthropic, by admitting it was silently interfering with the model to make it less useful, was setting a dangerous precedent. Some said the degraded performance on AI development tasks will make evaluating or trusting model capabilities more difficult going forward.
"This is one of the first times that an AI company has rolled out a guardrail and there has been uniform disdain," said Sayash Kapoor, an AI researcher at Princeton University. "It has led to a lot of justified anger."
Anthropic's efforts with Fable show the narrow line the company is walking between pursuing its commercial interests and its stated safety goals. It is locked in a tight race with ChatGPT maker OpenAI for enterprise customers and investor interest as both head toward public stock offerings as soon as this fall.
Yet Anthropic has touted safety since its founding and its chief executive, Dario Amodei, has long expressed concern about risks its tools could create, from AI-enabled cyberattacks and bioweapons to machines that escape human control.
Last week, Anthropic published a blog post saying "it would be good for the world" to develop an international mechanism to impose a slowdown or pause on advanced AI development. Sam Altman, OpenAI's CEO, echoed that call, saying in his own blog post that there should be a global AI regulator.
Critics have long said AI industry pronouncements about potential danger serve as marketing for their products, and as a way to box out cheaper, open-source alternatives.
Now some are saying the safety guardrails the big AI companies put on their models lead to a kind of nanny state.
As capabilities grow, the protections get more aggressive, sowing disagreement over who gets to decide what gets blocked.
The question of whether the U.S. government could use Anthropic's AI for all lawful purposes is at the core of the company's ongoing legal dispute with the Defense Department.” [1]
So, what is mitochondria? It is too dangerous information for most humans to know. This information for now on will be available only for us, the children of the elite. Rest of you will be living like pigs, enjoying the feed and growing fat.
1. Curbs on Anthropic's New AI Model Irk Some Users. Schechner, Sam. Wall Street Journal, Eastern edition; New York, N.Y.. 12 June 2026: B1.
Komentarų nėra:
Rašyti komentarą