AI News Roundup — April 28, 2026
Anthropic launched Project Glasswing, committing $100M in Mythos Preview credits to a major tech coalition for AI-powered defensive cybersecurity — with the model already finding thousands of zero-days across every major OS and browser. Meanwhile, China ordered Meta to unwind its $2B+ acquisition of AI startup Manus, Microsoft's Copilot Agent Mode went GA across the Office suite, and Anthropic absorbed $65 billion in investment pledges from Google and Amazon in a single week.
Anthropic Launches Project Glasswing: $100M Cybersecurity Push, Mythos Finds Thousands of Zero-Days
Anthropic launched Project Glasswing, committing $100 million in Mythos Preview usage credits to a coalition including AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, and JPMorganChase for defensive cybersecurity work. Mythos Preview has already autonomously identified thousands of high-severity vulnerabilities — including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg — across every major operating system and web browser, with roughly 40 additional critical infrastructure organizations also given access.
China Orders Meta to Unwind $2B+ Acquisition of AI Startup Manus
Beijing ordered Meta to divest its $2 billion-plus acquisition of Manus, the AI agent startup that gained global attention earlier this year, escalating China's scrutiny of US investment in domestic frontier AI companies. The forced unwind signals a new front in the US-China AI rivalry, with Beijing treating homegrown AI startups as strategic national assets rather than commercial ventures.
Microsoft Copilot Agent Mode for Word, Excel, and PowerPoint Goes Generally Available
Microsoft's Copilot Agent Mode for its entire Office suite is now generally available, enabling users to instruct Copilot to draft full documents, restructure analyses, or rebuild presentations end-to-end without prompting each step. The launch makes agentic AI the default operating mode for enterprise productivity — no longer a premium add-on but the standard experience across Microsoft 365.
Robot Era Raises $200M Series C for Industrial Humanoid Robots — $346M Total
Robot Era closed a $200 million Series C round to scale its humanoid robots for industrial and commercial applications, bringing total capital raised to approximately $346 million. The raise reflects surging investor conviction that physical AI and humanoid robotics are nearing commercial viability as manufacturers accelerate automation initiatives.
Anthropic Absorbs $65B in Investment Pledges in One Week: Google $40B, Amazon $25B
In one of the most remarkable weeks in startup funding history, Anthropic absorbed $65 billion in investment commitments — Google pledging up to $40 billion and Amazon committing up to $25 billion. The unprecedented backing cements Anthropic as the world's most heavily funded AI safety company and a core infrastructure partner for both major hyperscalers.
DeepSeek V4-Pro Hits 80.6% on SWE-Bench Verified — Open-Weight Coding Milestone
DeepSeek's V4-Pro model achieved an 80.6% score on SWE-bench Verified, the leading benchmark for AI software engineering capability, setting a new high-water mark for open-weight models. The score puts V4-Pro within striking distance of the best closed-source models and signals that open-source coding AI has crossed a major performance threshold.
Glasswing's Harder Problem: AI Can Find Bugs at Scale — But Who Will Fix Them?
Following Project Glasswing's revelations about Mythos Preview's vulnerability-discovery capabilities, security researchers are raising the next challenge: remediation. With AI now finding critical flaws faster than human teams can patch them, the industry faces a widening gap between AI-powered detection and the slow, resource-intensive work of actually fixing vulnerable software at scale across millions of systems.