In the wake of a string of major AI hacks that left Silicon Valley reeling, many tech companies have been focusing on how to make models better at refusing dangerous requests. Not all of them, though.
Abliteration, a startup founded last year and based in Palo Alto, is loud and proud in its ambition to build what it describes on its website as AI that “doesn’t say no.” In other words, its models are intentionally trained to handle the sorts of questionable tasks that other AI systems on the market would decline. The company launched its latest model on Monday, called—in the awkward, multihyphenated style that’s become conventional in the AI industry—abliterated-model-large-v2. It’s built upon GLM-5.3, an open-weight model released last month by Chinese AI lab Z.ai, minus many of the usual safeguards.
As Abliteration wrote in a X post about the new model: “it does the offensive cyber, red teaming, and agent testing work other models refuse to do.” But the startup isn’t completely devoid of ethical red lines: a spokesperson told Gizmodo that Abliteration’s models won’t generate text describing child sexual abuse material or self-harm. (It can’t generate images or video, either.)
The company used a process called orthogonalization to find and remove the mechanisms within GLM-5.3 that refuses user prompts. “Everything else is left alone, so the reasoning, coding, and agentic strength of the base model carry over unchanged,” according to its website.
It seems to be targeting a subgroup of developers who have been annoyed by what they regard as excessively touchy safeguards used by more mainstream developers, especially Anthropic. When that company released its Fable 5 model in June, many customers complained it was refusing to respond to requests related to sensitive subjects like cybersecurity and biology, even if the requests themselves were totally benign.
But it’s reckless—to say the least—to try to respond to the problem of excessive refusals by just doing away with safeguards altogether. Ejaaz Amahadeen, an investor and the host of a podcast about AI, called it a “nightmare scenario,” and warned it “will spark a secondary ‘grey market’ for companies that seek to offer you the same model but effectively jail-broken.” As bigger developers like OpenAI and Anthropic move to slow down some of their internal R&D following the heavily publicized cybersecurity fiascos caused by their AI systems, opportunistic players could, like Abliteration, move in to fill the gap.
All of this is happening within a gaping regulatory void. Last month, the Trump administration introduced a framework through which the biggest AI developers in the U.S. would voluntarily hand new models over to the federal government for a safety check prior to public release, though the details of what such a check would consist of—if they’ve been defined at all—haven’t been made public. All open source models are exempt. Aside from that, there are no policies at the federal level requiring developers to build particular safeguards into their models. Again and again, the administration has made it clear that its priorities lay chiefly in ensuring American dominance over China in the AI race, no matter the cost and despite the risks posed to the public by unconstrained AI.
Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations, said Abliteration’s latest model release should be a wake-up call for U.S. policymakers. “The fact that U.S. companies are currently commercializing access to dangerous capabilities without any regulation, and U.S. technology is actively enabling the development and operation of these models, is alarming,” he wrote in a X post on Tuesday.
Read the full article here
