Anthropic on Thursday admitted that it “made the wrong trade-off” when it released the Claude Fable 5 model a few days ago, according to multiple media reports. The announcement from Anthropic came after several researchers and developers began criticising the AI startup for limiting the capabilities of its AI model without informing users.
In case you aren’t aware, Fable 5 is the first Mythos-class AI model that Anthropic has released to the public. The company announced Mythos back in April but refused to release it publicly due to the cybersecurity risks the model posed.
With Fable 5, Anthropic announced that any requests pertaining to cybersecurity, biology or chemistry would be handed off to its Claude Opus 4.8 model. The company defended these restrictions by claiming that Mythos-class models can be used to develop cyberattacks or assist in biological research with potentially harmful applications.
However, researchers later found out that Fable 5 was also being quietly degraded when used for developing AI models, which would essentially prevent researchers and competitors from using Claude to build or improve rival AI systems. The company, however, did not notify users when their requests were being passed off to a less capable system.
Anthropic apologises for Fable 5’s safeguards:
Anthropic now says it made a mistake with Fable 5’s configuration. The company says it will now alert users whenever Fable 5 refuses a request or redirects them to a less capable model.
“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” the company told WIRED. “We made the wrong trade-off and we apologise for not getting the balance right.”
“Starting this week, flagged requests will visibly fall back to Opus 4.8. On the API, any flagged requests will return a reason for their refusal,” the company said in a statement to Business Insider.
The company reportedly said that the safeguards were introduced to address national security concerns and ensure that “foreign adversaries do not get ahead of the US in developing frontier chips and LLMs.” The company added that the majority of coding and machine learning work was unaffected by these safeguards.
The San Francisco-based AI startup had recently argued that the world should retain the option to slow down or temporarily pause the development of advanced AI models if safety research fails to keep up with advances in model capabilities.
How powerful is Fable 5?
In a blog post this week, Anthropic claimed that Fable 5 is not only its most powerful model released to date but also outperforms rival AI models from OpenAI and Google across a variety of benchmarks.
The company said that during early testing with Stripe, Fable 5 was able to compress months of software engineering work into days. The AI model is said to have completed a 50-million-line Ruby codebase migration in a single day that would otherwise have required a team of engineers working for more than two months.
Anthropic also said that Fable 5 excels at tasks that require vision capabilities, such as extracting information from scientific charts, understanding complex diagrams, and even recreating a web application’s source code using screenshots alone.
Source link
#Anthropic #apologises #hidden #safeguards #Mythosbased #Fable #model #wrong #tradeoff #Mint
