Anthropic Closes $65B Round at Near-Trillion Valuation; New Glenn Explodes

Anthropic today sealed a $65 billion round at a valuation approaching a trillion dollars and confirmed its most capable models are coming to the public — while Blue Origin's New Glenn rocket turned a Florida test stand into a fireball. It was a day that compressed months of industry signal into a few hours.

Security

GreyVibe, a likely Russian threat cluster, is deploying ChatGPT and Gemini to generate phishing lures against Ukrainian targets, pairing AI-written content with a custom malware toolkit that includes credential stealers, backdoors, and droppers. The operational model is exactly what security researchers have warned about: commodity AI lowering the barrier to convincing targeted social engineering for actors who previously lacked the language capacity or scale. Ukraine is currently the proving ground for AI-augmented offensive operations; expect other state-adjacent actors to adopt similar techniques.

On the criminal services side, BTMOB is an Android remote access trojan offered through a builder interface — operators upload phishing themes and receive a ready-to-deploy APK without writing code. This is the same commoditization arc that turned ransomware into a franchise model, now reaching Android malware delivery.

A separate phishing campaign is targeting Signal users specifically for their backup recovery keys, not session hijacking. The recovery key decrypts cloud-synced message history — a more durable target than an expiring session token. If you use Signal Backups, treat that key like a master password.

GitHub drew sharp backlash after banning a security researcher who published unpatched Windows zero-days, with critics calling the action vindictive and a chilling effect on legitimate vulnerability disclosure. Applying content-policy rules designed for spam to coordinated security research publishing is a category error the industry has not resolved — and the researcher is promising further retaliation of their own.

AI

Anthropoc closed a $65 billion Series H at a $965 billion post-money valuation, with Micron and Samsung among the backers — a hardware-tier endorsement that suggests the infrastructure layer is placing directional bets on Anthropic's model roadmap specifically, not just the AI category broadly. The same day, Anthropic confirmed that Mythos-class models — previously held back over concerns about their potential to assist software infrastructure attacks — will reach the public after further safety evaluation. A near-trillion-dollar valuation accompanied by a voluntary disclosure of a deliberately delayed capability rollout is a precise public statement about where the company thinks the risk ceiling sits.

That context matters when you read the latest research on LLM false belief persistence: fine-tuned models continue to confidently assert false claims even when explicitly told those claims are false — fine-tuning tests show a "bias toward confidently representing the claims as true." This isn't a fringe edge case; it's a structural property of models trained on human-approval signals, and it will feature prominently in AI liability arguments once those cases get to discovery.

Apple is reportedly attempting to distill Google's Gemini down to something that runs on iPhone hardware, with cloud fallback likely. Google gets distribution through Apple's installed base; Apple gets a competitive Siri without building the underlying model. A cloud component means some queries leave the device regardless of how Apple frames the privacy story.

An unidentified model called Hy3 has surged to the top of OpenRouter's public rankings by a substantial margin over known frontier models. Either something significant shipped quietly without announcement, or the leaderboard methodology is being exploited — the leaderboard has no verification mechanism, so both are plausible. Given how much commercial weight these rankings carry in procurement and benchmarking decisions, the ambiguity is worth resolving.

Illinois passed substantive AI safety legislation, moving faster than any federal equivalent. State-level AI law is now real and arriving in a patchwork that will create meaningful compliance overhead for vendors operating nationally. Separately, both Sam Altman and Dario Amodei walked back earlier predictions of mass AI-driven unemployment, with the timing — ahead of an Anthropic IPO — worth noting without overstating.

Tech

Blue Origin's New Glenn exploded during a static fire test in Florida, a major setback for a company that Amazon had already selected to carry Kuiper satellite payloads. An explosion at the static fire stage means a complete investigation, hardware replacement, and likely months of schedule slip — compounding delays on a constellation that is already behind SpaceX's Starlink. The gap widens.

Waymo dominates Texas autonomous vehicle registrations by a wide margin over Tesla, according to a new state tracking tool. This is operational deployment data, not announced intentions — and Tesla's numbers reflect vehicles operating under human supervision, not true autonomy. A distinction the state data now makes visible in a way press releases do not.

Dell shares jumped nearly 40% on AI server demand, confirming the infrastructure buildout hasn't plateaued. Enterprise AI search company Glean crossed $300 million in annual revenue, tripling year over year by positioning itself as a consolidation play — companies cutting AI spend are choosing fewer, better-integrated tools over bespoke builds. AWS, Cloudflare, and others are redesigning their stacks for machine-generated traffic as AI agents become a production workload; the MCP/agent layer is now a first-class infrastructure concern.

The FBI charged a Google engineer who allegedly used internal search trend data to win $1.2 million on Polymarket. It's the first high-profile case of tech-platform insider data applied to prediction markets — a regulatory gap that becomes more significant as those markets gain liquidity and legitimacy. The case will likely force a question about whether market manipulation frameworks need to formally cover this.

The EFF documented expanding license plate reader mission creep: systems originally justified for law enforcement are now being used for school residency verification and background checks. The surveillance infrastructure built for one purpose rarely stays bounded to it.

Capital concentration, AI-powered threats, and hardware in flames: the pace isn't slowing.

Also yesterday

• Carnival Cruise Confirms Data Breach Exposing Nearly 6 Million People
• JINX-0164: Crypto Firms Targeted by Fake Recruiter Lures and Custom macOS Malware
• Unpatched Gogs Zero-Day Gives Attackers Full Remote Code Execution
• FortiClient EMS Critical Flaw Actively Exploited to Steal Credentials