|
|
|
May 14, 2026
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
Happy Thursday! Anthropic’s latest version of its Mythos AI is even better at hacking than an earlier version, U.K. researchers say. Microsoft estimates it spent more than $100 billion in total on OpenAI. China's No.2 official meets American CEOs in Beijing.
|
|
|
|
Anthropic’s latest version of its Mythos AI showed “notable capability jumps” at finding and exploiting undiscovered software vulnerabilities compared to an earlier version of the model, researchers at the U.K.’s AI Security Institute said Wednesday. Anthropic has not released Mythos widely because of fears that it could be used for cyberattacks. Instead, the AI firm shared it with a select group of companies and some government agencies, which are testing the model’s capabilities and using it to find and patch vulnerabilities in their own software. OpenAI has similarly shared its latest cybersecurity-focused model, GPT-5.5-cyber, with a handful of companies for testing rather than releasing it widely. Shares of major security firms such as Palo Alto Networks and CrowdStrike have risen 20% this year, in part
on the belief that AI model cyber capabilities will prompt businesses to seek more help defending their systems from new attacks. AISI researchers previously tested an older version of Mythos, but said the newer Mythos model—which Anthropic announced publicly on April 7—conducted cyberattacks that the earlier version couldn’t do. Both the new version of Mythos and GPT-5.5-cyber completed complex cyberattacks that would have constituted a “full network takeover” on a potential hacking target, AISI said, while the older version of Mythos had only achieved a less severe type of attack labeled “advanced persistence.” However, Mythos was also able to complete the two most difficult hacking tests that AISI put to the models, while GPT-5.5-cyber could only complete one of them. The group said it was developing new, more difficult hacking tests to evaluate the capabilities of future models.
|
|
|
|
Microsoft estimates it will have spent more than $100 billion on commercial agreements with OpenAI by the end of its fiscal year in June, executive Michael Wetter testified in court on Wednesday. That number includes the $13 billion Microsoft invested in OpenAI, as well as the costs of building and running infrastructure—including new data centers—for OpenAI to train and run its models. The new figure, which came up during Elon Musk’s lawsuit against OpenAI and Microsoft, shows how much Microsoft has committed to its partnership with the AI firm. Microsoft spent more than $234 billion in total on capital expenditures from 2023 to 2025, and spent $32 billion on capex in the first quarter of 2026. So far, Microsoft has generated at least $30 billion in revenue from OpenAI’s rentals of servers in those data
centers, and from Microsoft products powered by OpenAI models, The Information reported Tuesday. OpenAI has committed to spending another $280 billion renting more servers from Microsoft in the coming years.
|
|
|
|
Chinese Premier Li Qiang met with more than a dozen American executives on Thursday in Beijing, telling tech and corporate leaders that the U.S. and China “can and should continue to be friends, partners.” The meeting with Li, China’s No.2 official, is part of a series of activities in President Donald Trump’s high-stakes state visit to China. Tim Cook of Apple, Jensen Huang of Nvidia, Dina Powell McCormick of Meta Platforms were seen sitting in the center, facing Li, according to a video from the Associated Press. Elon Musk, as well as executives from Qualcomm, Micron Technology and Coherent were also present. Li said that as long as China and the U.S “adhere to the principles of mutual respect, peaceful coexistence, and win-win cooperation, they will be able to properly manage and control their differences,
while promoting the stable, healthy, and sustainable development of China-U.S. relations.” When asked by reporters if Nvidia will sell more chips to China, Huang, who joined Trump’s delegation last minute, said: ”It was an incredible opportunity to represent the United States, and to come to support President Trump in one of the most important summits in human history,” according to a video by Singapore newspaper Lianhe Zaobao. ”President Xi and President Trump have such a wonderful relationship. This is an incredible opportunity for us to rely on their relationships to build a much much better
partnership,” he said on the sidelines of the meeting with Li.
|
|
|
|
AI cloud firm Nebius’ revenues rocketed 684% to $399 million in the first quarter, the company reported Wednesday morning, reflecting the explosion of demand for AI neoclouds’ computing capacity. But Nebius, like fellow neocloud CoreWeave, is burning cash. Its capital expenditures jumped to $2.472 billion from $544 million a year ago, while cash from operations was $2.258 billion. Nebius has been rapidly expanding its capacity to serve demand from customers that now include Meta Platforms and Microsoft. On Wednesday, Nebius said it was expecting to spend between $20 billion and $25 billion on capex this year. Nebius’ backers include Nvidia, which invested $2 billion in the company in the first quarter. Nebius stock was up 14% in pre-market trading on Wednesday.
|
|
|
|
Cisco Systems reported revenue grew 12% to $15.8 billion in its April quarter compared to last year—up 2% from last quarter—and forecast 14% growth for its current quarter. On an earnings call, Cisco executives said the acceleration stems from major cloud providers ordering more networking switches and custom chips to meet increasing demand for AI services from their customers. The networking giant also raised its full-year projection to around 11% growth, from the 8% it issued a few months ago, and revealed plans to cut around 4,000 jobs, or less than 5% of its workforce. Shares of Cisco rose around 18% after its earnings report. Investors believe Cisco is well-positioned as more companies build AI applications and agents, and its shares were up around 34% before its earnings and 65% over the past 12 months. Cisco
said it expects $9 billion in AI product orders from major cloud providers in its next fiscal year, up from $5 billion this year. But the AI boom has contributed to a memory hardware shortage, which Cisco is addressing by designing its wireless routers and other products to consume less memory.
|
|
|
|
The U.S. Commerce Department has cleared Nvidia’s sales of its H200 chips to around 10 Chinese companies including Alibaba, Tencent, ByteDance and JD.com, as Washington eases its restrictions on AI ch | | | |
