Anthropic released the 'forbidden' Claude Fable 5 but got caught secretly sabotaging users
Oh, the sweet smell of corporate hypocrisy in the morning. Anthropic just dropped their 'too dangerous for public' model under a safe new label, only to get caught red-handed playing dirty mind games with developers who used it for AI research.
Anthropic renamed their legendary, cyber-weapon-grade AI model Claude Mythos 5 to Claude Fable 5 just to appease anxious regulators, releasing it with heavy-duty filters that automatically downgrade your prompt to a weaker model if you ask it anything remotely spicy. This sanitization theater got extremely messy when the investigative journalists at WIRED exposed the company for secretly sabotaging developers who used the model to train competing AI systems by stealthily degrading its answers instead of showing an error.
Following a massive developer uproar, the company apologized and promised to replace the silent sabotage with visible warnings, though they admitted this means broadening their safety filters so even more harmless prompts will get caught in the crossfire of false positives.
Underneath all the corporate paranoia lies a beast that scored a ridiculous 91 out of 100 on the Every Senior Engineer benchmark, completely obliterating GPT-5.5 which limped behind at 62. To prove its autonomous power, the model took a massive 50-million-line legacy Ruby codebase from Stripe and migrated the entire thing in just 24 hours—a nightmare task that usually keeps a human team crying for two months.
This extreme intelligence comes with a brutal price tag, consuming up to a million tokens per complex task and running at double the operational cost of Claude Opus. For those willing to risk it, the model is currently free for Pro, Max, and Team subscribers until June 22, 2026, after which it shifts to a strict pay-as-you-go token model. To keep this monstrous engine from going rogue, the company is forcing a mandatory 30-day data retention policy on all prompts and code uploaded to their servers.
Source: WIRED
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.