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.
Anthropic has launched its first publicly available ‘Mythos-class’ AI model, Fable 5. Essentially, it’s the Mythos base model with some security restrictions implemented – namely that it will refuse user prompts related to cybersecurity, biotech, and AI model training. On the benchmarks, it’s pretty much head and shoulders above other public frontier models, and is particularly strong on long-horizon tasks. It’s also fairly expensive, at $10 per million input tokens and $50 per million output tokens. “Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage,” Anthropic said in its release announcement.
Neura Robotics has announced a Series C funding round targeting up to $1.4 billion in financing, with backing from Tether, Qualcomm Technologies, Amazon, Nvidia, Bosch, Schaeffler, and the European Investment Bank, among others. It’s an important signal for Europe, with Germany-based Neura planning to use some of the funding to open “Neura Gyms” that will serve as real-world training environments for its robots to learn to work alongside humans. The Financial Times reports that the deal values NEURA at about $7 billion, making it one of Europe’s most valuable AI startups.
The investment group KKR has launched a new AI infrastructure company called Helix Digital Infrastructure, backed by more than $10 billion in committed capital from the Kuwait Investment Authority, Nvidia, the power company Vistra, and KKR itself. The company will be headed by former Amazon Web Services CEO Adam Selipsky, and will generally concern itself with building out the infrastructure needed to support the AI boom, including designing and building data centers and securing their power sources. Helix will deploy Nvidia’s DSX AI infrastructure, which points to vertically integrated AI facilities optimized around Nvidia hardware and reference architectures.
Canada’s Privacy Commissioner Philippe Dufresne has determined that Grok’s generative image activities have violated Canada’s privacy laws by generating non-consensual sexualized deepfakes. The OPC report says Grok had generated more than 6,000 sexualized deepfakes per hour at one point, and that its output included the generation of sexualized images of children. In response, xAI and X Corp., which oversees the social media platform on which Grok interacts with users, disagreed with the OPC’s conclusions but committed to quarterly reports, third-party audits, and further safeguards.
Meanwhile, SpaceX just had its IPO, making its subsidiary xAI the first major frontier LLM company to go public (in a sense) – and making owner Elon Musk the world’s first trillionaire, at least on paper. It’s the biggest IPO in history, raising $75 billion at an IPO price of $135 per share. And the stock has enjoyed a bit of a pop in its first hours of trading, exceeding $175 per share at the time of writing. Among other things, SpaceX plans to use some of its newfound capital to help put AI data centers into orbit, where they can more easily reap the benefits of solar power (in theory).
The European Commission has published a Code of Practice on Transparency of AI-Generated Content under Article 50 of the EU AI Act, laying out some rules about detecting and labelling AI-generated content. The code distinguishes between “providers” – the AI companies, which must make sure that machine-readable watermarks are included in AI-generated content – and “deployers”, the entities using AI to alter media, and who also must mark their generations as manipulated or deepfake content. The Commission says the rules are meant to address “risks of deception and manipulation” and support “the integrity of the information ecosystem.”
The chatbot’s take: Fable 5 would like to correct the record.
–
By Alex Perala
