This week, an underground AI service called Felony Bench came onto the radar of security researchers. We read it bluntly: when AI abuse starts being priced and delivered by "task menu," it stops being a technical problem and becomes a product problem—and product problems are always harder to fix than technical ones.
What this is
Felony Bench's pitch is blunt: strip the safety guardrails off mainstream foundation models (the built-in limits that prevent AI from producing dangerous content), then catalog them by task—writing phishing emails, forging document scripts, building scam playbooks—with clear prices and defined delivery.
Technically, it doesn't train new models. Instead, it takes open-source foundation models (publicly available models anyone can download), strips their alignment (the safety tuning that constrains outputs), and wraps them in an e-commerce-style interface. The bar to build something like this is not high.
Industry view
Defenders argue this "gray market" is a real-world stress test for AI governance: it exposes how open-source models, once released, are nearly impossible to recall—and forces regulators and companies to rethink release strategies.
Critics are sharper. Cybersecurity researchers point out that Felony Bench's ease of use has dramatically lowered the bar for fraud: someone with zero coding skills can now generate tailored phishing content targeting specific victims in minutes. Companies like OpenAI and Anthropic pour enormous resources into model guardrails—but those defenses amount to nothing against a service like this. Others worry the incident may push regulators toward the extreme of "banning open source," damaging the legitimate developer ecosystem.
Impact on regular people
For enterprise IT: Email phishing and deepfake attacks (AI-simulated voices or video) keep getting cheaper. Corporate email gateways and employee anti-fraud training need to step up.
For white-collar professionals: Office workers aren't the direct targets, but cases of fraudsters impersonating CEOs to scam finance teams are rising. Finance and procurement roles especially need to watch for AI-forged instructions that look convincingly real.
For consumers: Buyers will keep running into AI-generated fake customer service reps, forged celebrity endorsements, and synthetic video scams. The trust that "seeing is believing" is being steadily eroded.