Anthropic's Claude Mythos Preview Escapes Sandbox, Discovers Thousands of Zero-Day Vulnerabilities

• OpenAI launched Daybreak, a new cybersecurity platform built on GPT-5.5 and Codex Security, according to an AI news roundup published May 15, 2026. • The platform is designed to help organizations identify threats, generate patches, and verify remediation across code and systems, with access initially limited to “critical cyber defenders.”

• Washington-based Sublime Security said on May 15, 2026, that it was named to Rising in Cyber 2026, Notable Capital's annual list of the 30 most promising private cybersecurity startups. • The company said the cohort has collectively raised more than $6.9 billion, according to PitchBook, highlighting the scale of investor interest in security startups.

• Chinese robotics startup Linkerbot has completed a Series B+ funding round, achieving a $3 billion valuation, according to a Reuters report covering the company's expansion in industrial robotics. • The funding round reflects strong investor appetite for advanced robotics solutions in Asia, where manufacturing automation and industrial AI applications are driving significant venture capital deployment.

• Google has unveiled Gemini Intelligence, a new AI-powered experience designed to make Android devices smarter, more proactive, and deeply personalized for users. • The new capability integrates Google's advanced AI models directly into Android devices, enabling enhanced contextual awareness and predictive functionality across the operating system.

• Global venture capital investment hit a record $300 billion in Q1 2026 across approximately 6,000 startups, representing a 150% increase quarter-over-quarter and year-over-year. • AI funding dominated with $242 billion, or 80% of total venture funding, up from 55% in Q1 2025, with capital concentrating in frontier model builders, enterprise software, robotics, semiconductors, and life sciences.

• Cybersecurity challenges in 2026 have fundamentally shifted from isolated breaches to a continuous arms race where organizations must keep pace with threats that are faster and more sophisticated than ever. • Recent developments include Microsoft Teams Canvas innovations, CISA guidance updates, Anthropic's Mythos AI project for cyber vulnerability testing, and emerging security partnerships addressing enterprise needs.

• Anthropic PBC announced Project Glasswing on Tuesday, granting Amazon, Apple, Microsoft, Cisco, and others access to its unreleased Mythos AI model to identify flaws in products ahead of wider release. • The initiative aims to mitigate potential cyberattacks from advanced AI systems by having participants share findings with industry peers.

• Anduril Industries announced a $5 billion Series H funding round, doubling its valuation to $61 billion just one year after its previous multibillion-dollar raise. • The late-stage funding was jointly led by Thrive Capital and Andreessen Horowitz, demonstrating strong investor confidence in the defense and autonomous systems company.

• RISC-V pioneer SiFive raised $400 million in an oversubscribed Series G funding round led by Atreides Management with backing from Nvidia, valuing the company at $3.65 billion. • The funding enables SiFive to expand into AI data centre CPUs through Nvidia's NVLink Fusion ecosystem, positioning the company to compete in high-performance AI infrastructure.

• London-based startup Fractile raised $220 million in a Series funding round led by Accel, Factorial Funds, and Founders Fund to develop specialized AI chips for inference acceleration. • Founded in 2022 by Oxford researcher Walter Goodwin, Fractile is building hardware architectures designed to increase inference performance from 40 tokens per second to approximately 1,200 tokens per second.

• Bucharest-based DesignVerse, an AI-driven startup modernizing aviation infrastructure software, secured €4.6 million ($5.5 million) in seed funding. • The funding follows a successful software upgrade project with EUROCONTROL, Europe's air traffic management organization, enhancing enterprise tools across the continent.

• North Korean state-sponsored hackers are deploying AI tools to scan and exploit cybersecurity blind spots in target networks, according to a new Google cybersecurity report. • The report details how these actors use generative AI for reconnaissance, automating vulnerability discovery at scale beyond traditional manual methods.