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.
Toronto-based AI firm Cohere has agreed to buy the German startup Aleph Alpha, with both firms framing the deal as a tech sovereignty play. Both firms have been working on LLM-based AI systems tailored for enterprise use, and while their models aren’t expected to ever reach the generalized intelligence of OpenAI and Anthropic models, they also don’t depend on American AI infrastructure. The deal comes after Canada and Germany struck a Sovereign Technology Alliance earlier this year, and Cohere said in its announcement that the combined entity would offer an “independent, enterprise-grade sovereign alternative” for organizations that want more control over how AI is deployed, governed, and hosted. Terms of the agreement were not disclosed, but it comes with a $600 million structured financing commitment from Schwarz Group.
The US National Security Agency is reportedly piloting Anthropic’s Mythos Preview model, which has been offered to select partners in a closed testing phase meant to help them adjust their cybersecurity posture given its enhanced hacking abilities. It’s a remarkable development given Anthropic’s recent and very public dispute with the Department of Defense over how its AI models could be used, which led to it being labelled a “supply chain risk” – a designation that Anthropic has been challenging in court. The NSA is a security agency within the DoD.
SK Hynix reported its strongest quarter ever, with revenue surpassing 50 trillion won for the first time and operating profit reaching a record 37.6 trillion won. The South Korea-based firm is an Nvidia supplier, and is riding the wave of demand for memory chips to support real-time inference workloads that require large volumes of DRAM and NAND. “Client requests for [HBM] chip supplies over the next three years already far exceeds our production capacity,” said Ki Tae Kim, SK Hynix’s head of HBM sales and marketing.
YouTube is officially rolling out its deepfake detection technology to Hollywood. The tech is designed to look for faces in uploaded videos that seem to match those of enrolled users. When a match is found, it’s sent to the user for review, and the user can then submit a likeness removal request. A pilot of this system had previously trialled it with video creators under the YouTube Partner Program, and it was expanded to government officials, journalists, and political candidates last month. Now, YouTube is working with major entertainment representatives including CAA, UTA, WME, and Untitled Management to refine the tool for Hollywood talent.
Neurotechnology subsidiary SkyBiometry has launched an AI infrastructure services business focused on hardware design, private bare-metal AI cloud operations, and Kubernetes-based model deployment. A growing number of organizations are going to need high-performance AI deployment environments without relying on a hyperscaler, and SkyBiometry is looking to fill that gap. Founded in 1990 and having since been involved in major biometric projects worldwide, Neurotechnology became unusually experienced in neural networks and high-stakes deployment, long before the current AI boom. “Our focus is on removing the infrastructure bottlenecks that hinder AI innovation,” said SkyBiometry CEO Mantas Kundrotas.
A Sony-made robot has proven to be competitive in table tennis. Dubbed “Ace”, the robot has defeated professional players, though it has lost some matches, too. The robot uses nine active pixel sensor cameras with Sony IMX273 image sensors to determine the ball’s 3D position while playing, and can track spin exceeding 9,000 revolutions per minute. Professional player Mayuka Taira, who lost to Ace in December, said the robot’s strengths are that “it is very hard to predict, and it shows no emotion.”
And finally: As detailed in this week’s Control Plane newsletter, members of the US House of Representatives were invited by the DHS to a closed session in which they were given access to some jailbroken AI models. Here are some of their reactions:
–
By Alex Perala
