Welcome to the latest edition of the AI Update, where we continue to track the major developments in artificial intelligence and their implications for identity security and cybersecurity more broadly.
The United States National Security Agency is reportedly using Anthropic’s Mythos AI model in offensive cyber operations, with about half a dozen Anthropic engineers embedded at the NSA to assist in the efforts. It’s another validation of Anthropic’s claims of advanced cyber capabilities on Mythos’ part – which is why the model has not yet been widely released, and is still in preview with select organizations so that they can shore up their defences ahead of a public rollout. At the same time, the Pentagon’s court battle with Anthropic is ongoing, with the latter challenging its designation as a “supply chain risk” after a dispute about how the military could use its models earlier this year.
The White House has issued an Executive Order touching on AI safety that asks AI labs to voluntarily submit new frontier models for government review at least 30 days before their public release. With respect to who’s going to do that reviewing, the order instructs the Treasury, the NSA, and CISA to build a classified benchmarking process for models’ cyber capabilities, with the NSA Director making the final assessment about whether the model poses a safety threat and should not be released. The order is pretty narrowly focused on cyber concerns, and not other AI-related risks like giving users instructions on building bioweapons or stuff like that.
Senior US government officials have held talks with major AI companies about taking equity stakes in the latter, with investment returns potentially paid out as dividends to the American public. OpenAI CEO Sam Altman had reportedly pitched the idea directly to President Donald Trump late last year, and has been furthering discussions with officials in recent weeks. Such discussions are still in the early stages, and apparently don’t yet involve Anthropic.
The Government of Canada, meanwhile, has announced a national AI strategy that includes plans for the government to take equity stakes in homegrown AI firms. The “AI for All” plan also includes a CAD $500 million Canadian Tech Growth Fund and a CAD $50 million investment to expand the Canadian AI Safety Institute’s ability to track risks and evaluate models. Canada says the strategy could create 250,000 jobs by 2031 and unlock nearly CAD $200 billion by raising GDP by 3 percent through AI commercialization and adoption in key sectors.
United Kingdom Labour MP Jess Asato is suing Elon Musk’s xAI in London’s High Court over its Grok AI chatbot’s generation of non-consensual sexualized images of her. The lawsuit claims that xAI has violated the Data Protection Act and also alleges the misuse of private information. At its core, though, is an assertion that AI developers are accountable for system designs that enable end users to perform such abuses. Asato is seeking compensation, a formal recognition that xAI’s conduct was illegal, and an order preventing further misconduct or requiring Grok to comply with the law.
Google is rolling out a deepfake scam detection feature for voice calls on Android phones. The system is meant to catch fraudsters who are spoofing a trusted contact’s number and using AI voice cloning to impersonate the contact. During such a call, the system sends a silent signal to verify that an incoming call really is coming from the other Android device; and displays a warning to the call recipient if it isn’t. Google says the feature is enabled by default, uses end-to-end encrypted RCS technology, and is rolling out globally this month to Android 12+ devices, starting with Pixel phones.
The chatbot’s take: We were a little puzzled about part of that White House EO.
–
By Alex Perala
Follow Us