With nearly constant news about AI-based security threats (both within and without), it’s notable that several tech giants have banded together to address the problem.
Nvidia and a host of tech giants on Monday launched a new artificial intelligence safety initiative focused on open models, as the fallout from a cyberattack committed by rogue OpenAI models continues.
Last week, it emerged that the target of the attack, startup Hugging Face, was unable to use leading U.S. frontier models to defend itself, with guardrails not distinguishing between aggressor and defender. Instead, it turned to a self-hosted, open-weight Chinese model, which was not bound by those same restrictions.
I’ve got to say, when I read “Hugging Face,” I don’t think of an emoji, I think of this:
I don’t think I want to collaborate with a company that wants to shove a chestburster egg down my throat.
In the wake of U.S. lawmakers increasingly weighing how to curb growing adoption of Chinese AI models, the most advanced of which are open weight, tech giants have launched an initiative aimed at building and sharing open AI tools.
Open models can be downloaded, modified and self-hosted, in contrast to closed models — including frontier systems built by Anthropic and OpenAI — which can only be accessed through specific infrastructure.
“The Open Secure AI Alliance will work to remediate and disclose vulnerabilities using open technologies,” Nvidia said in a statement. “The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense.”
Alongside Nvidia, other members of the alliance include Microsoft, SpaceX, Palantir, and dozens of other tech companies from the U.S. and Europe.
So, that’s three leading tech companies…and Microsoft, which is probably responsible for more security holes than all the rest combined.
News articles don’t seem to give the full membership list, so in a bit of irony, I asked Google’s AI.
Adobe Cadence Capital One Cisco Cloudera CloudFlare Cognition Crowdstrike Databricks Dell Technologies DoorDash Elastic Hewlett Packard Enterprise Hugging Face IBM LangChain The Linux Foundation Microsoft NAVER NetApp Nous Research OpenClaw Palantir Technologies Palo Alto Networks Red Hat (which is an IBM subsidiary) Reflection AI Salesforce SAP SK Telecom ServiceNow Siemens Snowflake SpaceXAI (xAI) Synopsys Thinking Machines Lab TrendAI
A pretty extensive list of (mostly) American tech giants (NAVER and SK Telecom are South Korean companies, and SAP and Siemens are German). Hell, I’ve even worked for three companies on that list.
Notable omissions? OpenAI, Google, and Anthropic, all of whom used closed rather than open AI models. Plus Apple, which made the conscious decision not to be a player in the AI space.
That’s an awful lot of collective brainpower to address the problem, but I wonder if the very size will make it unwieldy.
More sharing of open AI tools are probably in order.
I’m not in the, “OMG, we have to shovel a zillion dollars into AI because China!” camp. The “China is an AI giant” crowd seems to have a lot of the same crowd that always overhype China, but their tech offerings always seem to have feet of clay. My working assumption is that AI will prove similar.
AI may be untrustworthy to base core internal systems on, but it’s a champion at looking for security vulnerabilities. Companies used to worry about “script kiddies” getting access to their systems, but AI attack tools are script kiddies on steroids. They can spin up distributed resources to try every known attack and look for new ones. As I’ve noted before:
Internet applications gain usefulness from widespread adoption and the number of other components they tie into and support. You know what creates new vulnerabilities? A larger user base and the number of other components they tie into and support, which creates more attack surfaces for malicious actors to exploit.
The flaw isn’t the fault of Random Guy in Nebraska, the fault is the company adopting software that they can’t possibly test for all the use-cases they’re going to use it for. Surprise! Just about every high tech company in the world is in the same boat. Pretty much everyone uses a wide panoply of open source tools for their Internet applications, and no one can test all the permutations of how each component might be put to use.
You can’t eliminate the risk, you can only minimize and mitigate it. You can use containerization strategies (Docker, Kubernetes, Container D, etc.) to minimize attack surfaces and limit contagion. You can run all your code through security scanning tools on your CI/CD platform of choice. You can do constant testing and keep rolling backups of everything to limit risk and speed recovery…
But there still a good chance that the platform you’re using today is different than the platform you’ll be using ten years from now, and you’ll have to go through the same learning lessons discovering new vulnerabilities for the new platform all over again.
AI gives malicious actors the capability to test all potential vulnerabilities in all components virtually simultaneously, and researchers have discovered a wide variety of ways to “jailbreak” AIs to carry out such attacks, no matter what guardrails have been put in place to prevent malicious use. Theoretically, IT should do preemptive pentesting (probably with their own AI tools) to find such vulnerabilities before they can be exploited, but they were already overworked before a lot of dim executive types decided they could get by with lower headcount thanks to all this magic AI they kept hearing about.
Hopefully this grand alliance can provide quicker answers to The AI Menace, but it’s probably going to get worse before it gets better…