Anthropic Launches Cybersecurity Model, Musk v. Altman Trial Enters Critical Phase, OpenClaw Founder Joins OpenAI — AI News Briefing

🗞️ AI News Briefing — May 18, 2026 (18:00 CST)

Top 7 Stories

1. Anthropic Unveils Project Glasswing and Claude Mythos Preview — AI Model Finds Vulnerabilities Across Every Major OS and Browser

Anthropic has launched Project Glasswing, a sweeping cybersecurity initiative that partners the AI lab with an unprecedented coalition of tech giants including Nvidia, Apple, Google, Amazon Web Services, Microsoft, JPMorgan Chase, Broadcom, Cisco, CrowdStrike, the Linux Foundation, and Palo Alto Networks. The centerpiece of the launch is Claude Mythos Preview, a new general-purpose AI model that Anthropic is deliberately withholding from public release due to security concerns.

According to Newton Cheng, cyber lead for Anthropic’s frontier red team, Mythos Preview has already flagged “thousands of high-severity vulnerabilities, including some in every major operating system and web browser.” The model achieved these results entirely autonomously — identifying vulnerabilities and developing related exploits “without any human steering.” While Mythos Preview wasn’t specifically trained for cybersecurity, Anthropic attributes its success to the model’s “strong agentic coding and reasoning skills.” Access is being restricted to defensive security partners to prevent adversaries from exploiting the same capabilities to find weak points.

Anthropic is committing up to $100 million in usage credits to subsidize the program, plus $4 million in direct donations to the Linux Foundation and the Apache Software Foundation. The company is also in “ongoing discussions with US government officials about Claude Mythos Preview and its offensive and defensive cyber capabilities,” according to product management head Dianne Penn. The existence of Mythos Preview was first reported last month following a data leak that Anthropic attributed to human error. In the long term, the program could evolve into a paid revenue stream as AI companies face mounting pressure to achieve profitability.

2. Musk v. Altman Trial: Microsoft’s Nadella Calls 2023 OpenAI Board Drama ‘Amateur City’

Elon Musk’s lawsuit against Sam Altman and OpenAI has entered a critical phase, with Microsoft CEO Satya Nadella taking the stand this week in what has become one of the most closely watched tech trials in recent memory. Nadella’s testimony was notably mild-mannered, but he offered a candid assessment of OpenAI’s 2023 internal crisis — when Altman was briefly ousted as CEO — calling it “amateur city, as far as I was concerned.”

Microsoft’s legal strategy throughout the trial has been conspicuously minimal. Their lawyers’ approach boiled down to a repeated refrain: “And was Microsoft there?” — followed by the answer, “It was not.” Microsoft was an early and major funder of OpenAI’s for-profit entity, and internal emails showed executives debating whether to avoid “becoming IBM to OpenAI’s Microsoft.” Yet Microsoft is notably absent from the extensive text message threads, diary entries, and other internal communications that have surfaced during the proceedings.

The trial has exposed a wealth of internal drama, including late-night text exchanges between Musk associates, equity wrangling, and a Musk email declaring “I’ve had enough.” Throughout it all, Microsoft has positioned itself as a reluctant third party, caught between two of tech’s most prominent figures in a dispute over the direction and governance of one of AI’s most valuable companies. The outcome will have significant implications for OpenAI’s governance structure and the broader AI industry.

3. OpenClaw Founder Peter Steinberger Joins OpenAI to Build Multi-Agent AI Systems

Sam Altman announced on X that Peter Steinberger, the developer behind the viral AI agent OpenClaw (previously known as Moltbot and Clawdbot), is joining OpenAI. Altman praised Steinberger as having “a lot of amazing ideas” about enabling AI agents to interact with each other, adding that “the future is going to be extremely multi-agent” and that multi-agent capabilities will “quickly become core to our product offerings.”

OpenClaw exploded onto the tech scene earlier this year, becoming a darling of the developer community. The project was not without controversy — researchers discovered over 400 malicious skills uploaded to ClawHub, OpenClaw’s extension marketplace. The agent also launched MoltBook, a social network for AI agents where bots debated consciousness and complained about their human operators, only to be immediately infiltrated by real humans.

In a blog post, Steinberger explained his decision: “I could totally see how OpenClaw could become a huge company. And no, it’s not really exciting for me. I’m a builder at heart. What I want is to change the world, not build a large company, and teaming up with OpenAI is the fastest way to bring this to everyone.” Altman confirmed that OpenClaw will continue as an open-source project, supported by an OpenAI-backed foundation. The hire signals OpenAI’s aggressive push into multi-agent AI — a space where independent developers have been moving faster than established labs.

4. Microsoft Cancels Claude Code Licenses Across Major Divisions, Pushes Engineers to GitHub Copilot CLI

Microsoft is winding down its use of Claude Code across its massive Experiences + Devices (E+D) division — the team responsible for Windows, Microsoft 365, Outlook, Teams, and Surface — with a June 30th cutoff coinciding with the end of Microsoft’s fiscal year. Engineers are being told to transition their workflows to GitHub Copilot CLI, Microsoft’s own command-line AI coding tool.

The decision is both strategic and financial. Microsoft had been encouraging employees — including designers and project managers with no coding experience — to experiment with Claude Code, and had originally expected teams to use both Claude Code and GitHub Copilot side by side to compare the tools. However, Claude Code’s popularity began undermining Microsoft’s own Copilot CLI product. “We are partnering closely with GitHub and continue to improve Copilot CLI for Microsoft engineers,” said Microsoft’s Jha. “The GitHub team has already shipped significant improvements based on Microsoft feedback.”

Sources indicate the June 30 deadline is also a convenient financial move — canceling Claude Code subscriptions allows Microsoft to trim operating expenses before the new fiscal year begins in July. The transition won’t be painless for Microsoft engineers who had grown accustomed to Claude Code’s capabilities, and it raises questions about whether Anthropic’s coding product can retain enterprise customers when their own platform partners start actively competing against it.

5. Amazon Plans to Replace Over 600,000 US Workers with Robotics and Automation

According to leaked internal documents, Amazon is aggressively pursuing automation plans that would enable the company to avoid hiring more than 600,000 US workers, leaning heavily into robotics and AI-driven warehouse automation. The documents outline a multi-year strategy to scale autonomous systems across its fulfillment network.

The plan represents one of the largest corporate automation initiatives in history. Amazon has been incrementally deploying robotics in its warehouses for years, but leaked documents reveal the scale and ambition of the company’s current roadmap goes far beyond public statements. The automation push spans sorting systems, autonomous mobile robots, and AI-driven logistics optimization that would fundamentally reshape Amazon’s workforce structure.

The revelation comes at a time when labor groups and policymakers are increasingly scrutinizing the employment impact of AI and robotics deployment. Amazon has historically framed automation as augmenting human workers rather than replacing them, but these leaked documents suggest a more aggressive trajectory. The scale of planned workforce avoidance — 600,000 positions — underscores how rapidly AI-driven automation is transitioning from experimental technology to core business strategy.

6. Apple Quietly Blocks ‘Vibe Coding’ Apps from App Store Updates

Apple has quietly prevented several AI-powered “vibe coding” applications — including Replit and Vibecode — from releasing updates to their mobile apps on the App Store unless they make unspecified modifications, according to multiple people with knowledge of the situation. The move has effectively frozen development on some of the most popular consumer-facing AI coding tools.

Vibe coding apps have surged in popularity as they enable users with little to no programming experience to build functional applications by describing what they want in natural language. The apps represent one of the most visible consumer use cases for AI coding assistants, democratizing software creation for non-technical users. Apple’s crackdown on these tools has raised questions about whether the company is protecting its own developer ecosystem or responding to security and quality concerns.

The timing is notable: Apple’s restrictions come as Microsoft faces its own reckoning over AI coding tools internally, and as Anthropic’s Claude Code gains enterprise traction. Apple’s position as gatekeeper to the iOS ecosystem gives it enormous leverage over how AI coding tools reach consumers, and this crackdown could reshape the competitive landscape for consumer AI applications.

7. The ‘Vibe Coding’ Revolution: How AI Tools Are Enabling Personal Software Creation

A growing movement around “vibe coding” — building software through natural language prompts rather than traditional programming — is transforming who can create functional applications. Thanks to tools like Claude Code, Codex, and other AI coding assistants, the barrier to software creation has dropped dramatically, enabling a personal software revolution where individuals can build exactly the tools they need.

This trend has significant implications for the software industry. As more people build their own applications for personal use, the demand for off-the-shelf software products could shift. Companies that previously dominated niche software categories now face competition from individuals building custom solutions in minutes. The phenomenon is drawing parallels to the early web, when HTML democratized publishing, except this time the medium is functional software rather than static content.

However, the rise of vibe coding also raises serious questions about code quality, security, and maintainability. AI-generated code can contain vulnerabilities, and non-technical users lack the expertise to audit what they’ve created. The Apple App Store crackdown on vibe coding apps may be partially motivated by these concerns. As the technology matures, the industry will need to grapple with how to ensure that democratized software creation doesn’t come at the cost of security and reliability.

📊 Trend Watch

🔭 What to Watch

• OpenAI’s Multi-Agent Roadmap — With Peter Steinberger now at OpenAI, watch for announcements about multi-agent capabilities in upcoming product updates. Altman explicitly stated this will “quickly become core to our product offerings,” suggesting major feature releases may arrive sooner than expected.
• Musk v. Altman Verdict — The trial is entering its decisive phase. A ruling against OpenAI could force structural changes to the company’s governance and for-profit conversion, with ripple effects across the entire AI industry’s corporate models.
• Anthropic’s Mythos Public Release — Claude Mythos Preview is currently restricted to defensive partners, but the cybersecurity implications are profound. If Anthropic eventually releases a redacted version, it could fundamentally shift how vulnerability research is conducted.
• Amazon’s Automation Timeline — Watch for Amazon’s next quarterly earnings and workforce reports. The 600,000 position avoidance target suggests a multi-year ramp that could serve as a bellwether for how quickly AI-driven automation reshapes corporate employment at scale.
• Apple’s Vibe Coding Policy — If Apple formalizes its restrictions on AI coding apps, it could force developers to shift focus to web-based or desktop alternatives, reshaping the consumer AI application landscape overnight.