|
|
|
Jun 03, 2026
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
Happy Wednesday! Anthropic expands its Project Glassing initiative to let 150 more organizations use its Claude Mythos model for cybersecurity testing. Meta adds limits to its employee tracking tool after staff backlash. SpaceX is pushing the banks working on its IPO to cut fees.
|
|
|
|
Anthropic is quadrupling the number of companies allowed to use its Claude Mythos model for cybersecurity testing under its Project Glasswing initiative. The company announced Tuesday that the project will be expanded to 150 new organizations, from the roughly 50 that were included when it launched in April. Anthropic’s partners use the cyber-capable model, which hasn’t been more widely released yet, to find security vulnerabilities in their software code and patch them before cyberattackers exploit them. “What each partner has in common is that a successful attack on their codebase could be catastrophic,” Anthropic wrote in a statement. “For most partners, we estimate that a major attack could affect more than 100 million people, with important ramifications for both global and national security.” The organizations span more than 15 countries and several industries, including power, water and healthcare. Anthropic has faced criticism for limiting access to Mythos to a narrow group of companies. It has said it aims to release the model publicly but AI companies, including itself, haven’t developed measures to prevent their models from assisting with cyberattacks, in part because it is difficult to distinguish a harmful cyber task from a benign one.
|
|
|
|
Meta Platforms is scaling back elements of its employee tracking tool after staff raised concerns about the tool, The Information reported. A Meta executive told employees in the memo that the company is updating its Model Capability Initiative, or MCI, launched in April to help train its AI models, by strengthening privacy protections, enabling some staff to request exemptions and giving employees an option to pause the tracker on their computers for 30 minutes. Meta told staffers that while privacy safeguards were introduced when MCI launched, employees had raised concerns about battery life, personal data on work devices and having more control over when screen data is collected. The changes come after some Meta employees pushed back against MCI following its rollout.
|
|
|
|
SpaceX is negotiating to pay the banks working on its potentially record-breaking initial public offering a smaller than normal slice of the proceeds even by mega-IPO standards, Bloomberg reported on Tuesday. SpaceX is aiming to raise about $75 billion, more than any other IPO in history, The Information has previously reported. Banks generally charge fees of 4% to 7% of capital raised on smaller IPOs, though that percentage typically shrinks with mega listings. SpaceX is seeking a less than 0.75% fee, Bloomberg reported, while previous mega listings have usually commanded at least 1%. Even if SpaceX winds up getting the fee structure it wants, the banks would still earn around $500 million, marking a collective windfall for Wall Street. Banks may also receive fees on top of the 0.75% if the IPO goes well, Bloomberg reported. SpaceX publicly filed its IPO paperwork in late May and is due to start trading as soon as next week. The company has nearly two dozen banks working on the IPO led by Goldman Sachs and Morgan Stanley.
|
|
|
|
Microsoft on Tuesday revealed new AI software geared towards developers and corporate customers, including homegrown AI models and new tools that aim to let corporate customers develop AI agents that can draw on internal company data to automate workplace tasks. The announcements come as Microsoft is aiming to woo more developers to use its AI-powered software, a key plank of its effort to bolster GitHub and its Azure cloud platform against rising competition from the likes of OpenAI, Anthropic, and Google. Confirming an earlier report in The Information, Microsoft announced seven new homegrown AI models, including one model specialized for coding, which will be available in Microsoft’s GitHub Copilot AI coding tool as a cheaper alternative to models from OpenAI and Anthropic, Microsoft said. Other AI models unveiled Tuesday are specialized for reasoning, transcription, generating images, and speech. Microsoft also debuted a new AI agent product called Microsoft Scout, combining the popular open source OpenClaw agent software with customers’ internal company data, such as their Teams and Outlook messages, to work on long-running tasks autonomously within companies. For instance, a company can use Scout to keep tabs on employees’ meetings and automatically send meeting attendees prep notes gleaned from their emails, or suggest ways to reschedule meetings when there are conflicts, Microsoft said. And Microsoft shared more details about how businesses can connect their AI agents to internal data stored within their other Microsoft apps, such as Office 365 or Fabric data software. Microsoft will sell access to the data through products dubbed Microsoft IQ, which will be available as application programming interfaces that let AI agents pull data directly out of those applications to automate tasks within a company’s IT systems. The release echoes similar data products for AI agents released by the likes of Salesforce and Snowflake in recent months. The company also revealed new cybersecurity tools that help companies track and manage agents, and tools that direct AI agents to find and fix security bugs in companies’ systems. Finally, Microsoft unveiled new tools geared towards app developers, including a desktop app version of its GitHub Copilot AI coding tool and a new tool called Project Rayfin that lets developers quickly spin up websites that they code using AI tools. Microsoft also pitched the new Surface RTX Spark laptop powered by Nvidia processors that it announced on Sunday as a device geared towards developers.
|
|
|
|
Shopify said Tuesday that it would expand its share buyback authorization by $3 billion to a total authorization of $5 billion, becoming the latest software firm to unveil expanded share repurchase plans as AI fears have battered stock prices. Several software firms have implemented or expanded share buybacks as worries about AI disruption have sunk their share prices. In March, Pinterest said that activist investor Elliott Management had agreed to invest $1 billion in share repurchases, for a total of $2 billion in buybacks during the first half of this year. Salesforce and ServiceNow have also ramped up their buybacks. Shopify’s share price is down more than 25% so far this year. As of June 1, the company has repurchased around $1.45 billion so far this year.
|
|
|
|
| |
