London, April 20, 2026, 09:44 BST
Regulators in Australia and South Korea put Anthropic’s Mythos model under closer watch on Monday, making the cyber-capable AI system the clearest flashpoint in a wider day of artificial intelligence news spanning banks, chips, data centres and robotics. Australia’s securities watchdog said it was “closely monitoring” developments with peer regulators, while South Korea’s Financial Supervisory Service held a meeting with financial firms to review Mythos-related risks. Reuters
The urgency is that the latest AI systems are no longer just writing text or code on command. They are moving toward agentic AI — systems that can plan tasks and take actions with less direct human input — and that raises a harder question for banks and governments: whether the same tools that find software flaws can also help attackers exploit them.
Anthropic has tried to frame Mythos as a defensive project. Its Project Glasswing gives selected partners access to Claude Mythos Preview to find and fix vulnerabilities in critical software, and the company said launch partners include major cloud, chip and cybersecurity firms. Anthropic has committed up to $100 million in usage credits and $4 million in donations to open-source security groups, it said.
The chip trade moved in the same direction. Morgan Stanley said agentic AI could shift more spending toward central processing units, or CPUs, the general-purpose chips that coordinate computing tasks, and memory, even as demand remains strong for graphics processing units, or GPUs, the chips that dominate AI training and inference. “As AI transitions from generation to autonomous action, the computing bottleneck is shifting towards CPU and memory,” the firm wrote. Reuters
Google, meanwhile, is in talks with Marvell Technology to develop two new AI chips, The Information reported, in a sign Alphabet wants more control over the hardware used to run models. One chip would be a memory processing unit designed to work with Google’s tensor processing unit, or TPU, its in-house AI accelerator; the other would be a new TPU for running AI models. Reuters said it could not immediately verify the report and that Google and Marvell did not immediately respond to requests for comment.
The infrastructure story is still drawing capital. Australia’s NEXTDC said it planned to raise A$1.5 billion, or $1.07 billion, to speed the rollout of its S4 Sydney data centre, while contracted utilisation rose about 60% to 667 megawatts. CEO Craig Scroggie called it a “unique opportunity” to expand contracted capacity and de-risk Western Sydney developments before possible private-capital partnerships. Reuters
Suppliers around the chip chain are seeing the pull. British engineering firm Renishaw raised its 2026 revenue and profit forecasts, citing strong demand from semiconductor and electronics manufacturing equipment customers, with Reuters noting demand from AI data centres and autonomous vehicles. The company now expects adjusted pretax profit of 145 million pounds to 165 million pounds, up from a prior range of 132 million pounds to 157 million pounds.
In Europe, the regulatory fight sharpened. German Chancellor Friedrich Merz said industrial AI needed more freedom than consumer AI under EU rules, arguing that he would push to ease what he called a regulatory burden and, where possible, exempt industrial AI from a “regulatory straightjacket.” Germany wants to catch up with the United States and China in AI while using the technology to lift productivity and lower costs. Reuters
The China angle came from the physical world. Dozens of Chinese-made humanoid robots ran in a Beijing half-marathon, with the winning robot developed by Honor finishing in 50 minutes and 26 seconds. Reuters reported that more than 100 teams took part, up from 20 last year, and that nearly half the robots used autonomous navigation rather than remote control.
But the day’s headlines also showed where the risks sit. Regulators may slow or narrow access to powerful models if financial-system or cyber risks look unresolved; chip talks can fall apart before production; and data-centre economics still depend on power, equipment, customers and planning rules. Bloomberg reported that Siemens CEO Roland Busch warned the company would prioritize AI investment in the United States and China if the EU did not adapt its rules, another sign regulation is becoming a location decision, not just a compliance matter.
Taken together, the AI technology news today was less about a single product launch than control points: who can use frontier models, who builds the chips, who finances the data centres and who writes the rules. That is where the next phase of the artificial intelligence market is being fought.
