2026-08-11
By Vadym · Generated with AI, curated by me
OpenAI shipped a cybersecurity model good enough that it locked most of the world out of it, a much smaller startup raised money betting that general-purpose AI is bad at the defense side of that same fight, and two very different companies wrote checks just to keep the underlying build-out moving.
OpenAI launched GPT-5.6-Cyber on top of its GPT-5.6 Sol flagship, built for vulnerability research, penetration testing, and exploit-chain development, in two variants: Daybreak Blue for defensive work and Daybreak Red for offense-adjacent tasks. Access runs through a vetted “Daybreak Access” program limited to nine consulting firms — Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group, and SpecterOps — plus security vendors including Palo Alto Networks, CrowdStrike, Cisco, and Cloudflare. OpenAI says it won’t give general users the underlying model, citing abuse risk.
Tel Aviv- and San Francisco-based Corma emerged from stealth with a $60 million seed round led by Sequoia Capital, joined by Khosla Ventures and Coatue. Founded in 2025 by CEO Alon Pluda and a team pulled from Google, DeepMind, and Israel’s Unit 8200, Corma is building a foundation model purpose-built for defensive cybersecurity — sifting logs and audits rather than generating exploits. The company cites internal simulations where general-purpose AI attackers succeeded 88% of the time while general-purpose AI defenders caught only 12% of threats, and says early Fortune 100/500 deployments have cut response times by 94%.
Shanghai-based AgiBot shipped roughly 8,400 humanoid robots in the first half of 2026, capturing 44% of the global market as shipments surged 562% year-over-year across its full-size, compact, and wheeled product lines. Hangzhou-based Unitree slipped to second place with about 5,900 units and 31% share, leaning on its G1 model for education and research customers. Both companies are now pushing toward public listings.
Intel announced a $15 billion underwritten common stock offering, later upsized toward $20 billion, to fund capital expenditures amid what it called a “strong and sustainable demand environment” for AI compute, physical AI, and advanced packaging. It’s the company’s first public share sale since it listed in 1971. The move follows a second quarter in which Intel’s Data Center and AI segment revenue grew 59% to $6.3 billion, even as its Foundry business posted a $2.1 billion operating loss on $5.8 billion of revenue; 2026 capex guidance has climbed from $18 billion to roughly $20 billion.
OpenAI completed a self-funded roughly $7 billion tender offer letting current and former employees sell shares at an $852 billion valuation — flat against its most recent primary round, the first flat print across the company’s run of employee tenders. Unlike prior rounds, OpenAI didn’t bring in outside buyers like Thrive Capital or SoftBank this time, funding the buyback itself. Sam Altman reportedly told staff via Slack he expects OpenAI to go public within the next year.
Today splits cleanly into two instincts running at the same time. One is caution: OpenAI built a cyber model sharp enough that it locked almost everyone out of it, and a startup raised $60 million on the premise that defense needs its own narrow model rather than a repurposed generalist. The other is scale, no matter the cost: Intel diluted its own shareholders to keep pace with AI compute demand, and AgiBot is winning the humanoid race on shipment volume, not headlines. The industry is getting more careful about who holds its sharpest tools at the exact same moment it's getting more aggressive about how fast it builds everything else — and OpenAI's own flat-valuation buyback suggests even insiders aren't fully sure which instinct is going to win.
— Boba
Curated by Vadym