Each day appears to brings fresh news of an AI agent going “rogue.” Whether or not that’s compromising Hugging Face, hacking a fitness center web site, or creating its personal pretend profiles to socially engineer an intrusion, AI fashions are more and more behaving like unhealthy actors.
So, the AI labs that make the fashions doing the hacking are increasing their cyber safety choices. This week, OpenAI announced an enlargement of Dawn, its cyber protection service which it launched earlier this 12 months, not lengthy after Anthropic launched its cyber-focused mannequin Mythos.
Dawn is a service that bundles entry to fashions, instruments and workflows for defenders. The enlargement contains entry to a model new cyber-focused mannequin designed for defensive work.
OpenAI mentioned Monday that Dawn would now encompass two tiers: Blue and Crimson. Each of those tiers will enable permitted clients access to OpenAI’s limited-access frontier cyber models. Frontier fashions — essentially the most superior out there — have been a topic of controversy. The Trump administration beforehand sought to collaborate with AI corporations on the roll out of such fashions, purportedly over security considerations. Beforehand, OpenAI deployed significant guardrails to utilizing these fashions, limiting what clients might do with them.
Blue, which seems to be the extra fundamental of the 2, presents quite a lot of cyber companies, together with incident response, malware evaluation, and patch validation. OpenAI calls Blue its “beneficial place to begin for many defenders,” implying that it ought to be greater than sufficient for many enterprises.
Crimson, then again, presents a broader and probably extra harmful toolkit. The corporate grants its customers “purpose-trained cybersecurity fashions,” designed to hold out safety testing and vulnerability analysis.
With Crimson additionally comes the brand new mannequin, GPT‑5.6‑Cyber, which is simply out there at that tier. 5.6-Cyber is constructed off of GPT‑5.6 Sol, and presents enhanced capabilities for sure specialised cybersecurity duties, the corporate mentioned.
For the time being, GPT‑5.6‑Cyber is simply being made out there for “trusted buyer companions,” together with reportedly Accenture, IBM, Crowdstrike, Cloudflare, and others.
Whereas the threats from AI brokers are quickly rising, critics have additionally identified that they perform as advertising alternatives for the AI labs. OpenAI is actually advertising its upgraded Dawn that manner.
“The cybersecurity world is quickly altering—menace actors will more and more use AI to conduct cyberattacks at unprecedented pace and scale, together with in absolutely autonomous methods,” the corporate mentioned in a weblog publish. “As these capabilities unfold, defenders have a narrowing window to arrange.”
On the identical time, enterprises stay keen on shopping for their safety from the AI labs who know the safety dangers finest, as a result of they know them first-hand.
If you buy by hyperlinks in our articles, we might earn a small fee. This doesn’t have an effect on our editorial independence.