Ever since the GenAI wave of technology kicked off at the end of 2022, we have been warned that superintelligent AI is only a few years away. Anthropic’s Fable and Mythos models may be that turning point.
After Fable, the lighter model with more restrictions, was released to the general public on the 9th June, it was promptly suspended by the US government due to security concerns. This is a model that received an overwhelmingly positive review on Salesforce automation use cases, including Flow and Apex.
Fable and Mythos Security Concerns
Mythos is the overarching model that can apparently find and penetrate vulnerabilities in systems that have been live for decades, with Salesforce Architects warning against the threat of this model.
Mythos 5 has only been released to a select number of companies within the US due to concerns that it could be used maliciously by hackers, whilst Fable 5 was released to the general consumer, with strong safeguards to make it safer for general use.
In addition, OpenAI’s new GPT-5.6 model is going through a similar process with the Trump administration, over similar security concerns.
There have been concerns over the past year that AI Lab CEO’s such as Dario Amodei of Anthropic or Sam Altman of OpenAI have been scaremongering with their language to increase the perceived value of their own products.
Catchy soundbites such as “AI is going to wipe out the majority of white-collar jobs”, or that their latest models are “too powerful to release to the public”.
But with the US government, as well as Amazon CEO Andy Jassy (a large partner of Anthropic’s), apparently triggering the Trump administration’s involvement, it makes you question just how much of this is marketing vs reality.
Fable 5 Enabled
Nearly three weeks after the US Department of Commerce demanded Anthropic remove Fable 5 for any foreign national (before eventually removing it for everyone), they will slowly be enabling it across the globe from today, Wednesday 1st July.
Amazon’s report showed a jailbreak scenario, where Fable 5 was able to identify vulnerabilities and then produce code to exploit them.
Anthropic has since announced that Fable 5 has been updated to block the specific technique 99% of the time, and route dangerous requests to its previous model, Opus 4.8.
Impact on Salesforce Development
It’s safe to say that AI Development has received mixed reviews within the Salesforce ecosystem. While there are some vocal champions, AI coding slop, hallucinations, and an uncertain feeling about how this technology is going to be used and affect job roles has left many uneasy.
We are seeing overwhelming evidence at SF Ben that the developer role – once one of the most in-demand roles within the ecosystem – is slowly being eroded. You can see this in our latest architect survey, where AI Specialists are now being prioritized in the team over developers, and 10K’s Advisors report showing a 12% decline in demand YoY.
Our resident Flow master, Tim Combridge, tested Fable 5 before its restriction, and the results were extremely impressive over three use cases…
- Challenge 1: Build a Flow from scratch: Tim asked Fable to build an Autolaunched Flow with input/output text variables and a Get Records element to fetch an Account by ID and return its name. Fable produced it within minutes, and it deployed cleanly the first time, succeeding where earlier off-the-shelf models had failed, and notably doing what Agentforce Vibes (a Salesforce-trained tool) couldn’t.
- Challenge 2: Smart-clone a Flow: He gave a deliberately vague instruction to clone an existing Contact-retrieval Flow into one that fetched Opportunities instead, without even naming which Flow to copy. Fable worked out the right Flow, built the new version correctly including the fault path, and updated the “Contact” references to “Opportunities” in the descriptions.
- Challenge 3: Convert Flows to Apex: Tim asked Fable to merge the two Flows into a single invocable Apex class with test classes passing 100%, embedding a trap since you can’t have two @InvocableMethod annotations in one class. Fable adapted by writing one scalable method with the object and fields as input parameters, passed all tests, and even corrected a mistake Tim had accidentally left in his own Flow.
Final Thoughts
The battle between AI spend and the value companies are receiving rages on, with Accenture recently reporting that 90% of companies are yet to see a financial benefit from GenAI spend.
But we are still very much in the early innings; whilst AI might make us more productive as individuals, companies are not seeing the benefit in their bottom line.
But it’s safe to say we are entering a new era of models so powerful that governments have to get involved. Stay safe out there!
