Google brings generative AI to cybersecurity

There’s a brand new pattern rising within the generative AI area — generative AI for cybersecurity — and Google is amongst these seeking to get in on the bottom ground.

On the RSA Convention 2023 immediately, Google announced Cloud Safety AI Workbench, a cybersecurity suite powered by a specialised “safety” AI language mannequin known as Sec-PaLM. An offshoot of Google’s PaLM mannequin, Sec-PaLM is “fine-tuned for safety use circumstances,” Google says — incorporating safety intelligence comparable to analysis on software program vulnerabilities, malware, menace indicators and behavioral menace actor profiles.

Cloud Safety AI Workbench spans a spread of latest AI-powered instruments, like Mandiant’s Risk Intelligence AI, which is able to leverage Sec-PaLM to seek out, summarize and act on safety threats. (Recall that Google bought Mandiant in 2022 for $5.4 billion.) VirusTotal, one other Google property, will use Sec-PaLM to assist subscribers analyze and clarify the habits of malicious scripts.

Elsewhere, Sec-PaLM will help clients of Chronicle, Google’s cloud cybersecurity service, in looking out safety occasions and interacting “conservationally” with the outcomes. Customers of Google’s Safety Command Heart AI, in the meantime, will get “human-readable” explanations of assault publicity courtesy of Sec-PaLM, together with impacted belongings, really useful mitigations and danger summaries for safety, compliance and privateness findings.

“Whereas generative AI has not too long ago captured the creativeness, Sec-PaLM is predicated on years of foundational AI analysis by Google and DeepMind, and the deep experience of our safety groups,” Google wrote in a weblog publish this morning. “We’ve solely simply begun to appreciate the ability of making use of generative AI to safety, and we sit up for persevering with to leverage this experience for our clients and drive developments throughout the safety group.”

These are fairly daring ambitions, significantly contemplating that VirusTotal Code Perception, the primary instrument within the Cloud Safety AI Workbench, is just out there in a restricted preview in the meanwhile. (Google says that it plans to roll out the remainder of the choices to “trusted testers” within the coming months.) It’s frankly not clear how properly Sec-PaLM works — or doesn’t work — in follow. Positive, “really useful mitigations and danger summaries” sound helpful, however are the solutions that significantly better or extra exact as a result of an AI mannequin produced them?

In spite of everything, AI language fashions — irrespective of how cutting-edge — make errors. They usually’re inclined to assaults like immediate injection, which might trigger them to behave in methods their creators didn’t intend.

That’s not stopping the tech giants, after all. In March, Microsoft launched Safety Copilot, a brand new instrument that goals to “summarize” and “make sense” of menace intelligence utilizing generative AI fashions from OpenAI together with GPT-4. Within the press supplies, Microsoft — much like Google — claimed that generative AI would higher equip safety professionals to fight new threats.

The jury’s very a lot out on that. In reality, generative AI for cybersecurity would possibly change into extra hype than something — there’s a dearth of research on its effectiveness. We’ll see the outcomes quickly sufficient hopefully, however within the meantime, take Google’s and Microsoft’s claims with a wholesome grain of salt.