OpenAI Battles Anthropic for Cyber AI.

TL;DR

OpenAI's cybersecurity entry: A restricted-access model is being finalized via its Trusted Access program, directly paralleling Anthropic’s Glasswing deployment strategy.
OpenAI's ChatGPT Pro tier: A new $100/month plan offers 5x Codex usage, with exclusive GPT-5.3-Codex-Spark access, aligning precisely with Anthropic’s Max tier pricing.
Meta's CoreWeave expansion: An additional $21B commitment through 2032 raises Meta's total investment to $35B, securing early access to NVIDIA's Vera Rubin platform.
OpenAI's ad revenue forecast: Internal investor projections target $100B by 2030, initiating a direct market confrontation with Google and Meta’s established advertising frameworks.
OpenAI's liability shield initiative: The company supports an Illinois bill capping liability for "critical harms" — 100+ deaths or $1B+ in damage — contingent on maintaining a safety protocol.

Lead Story: OpenAI Enters the Cybersecurity Arms Race

OpenAI is finalizing a cybersecurity product featuring "advanced hacking and defensive capabilities," slated for restricted release to a select group of partners, Axios reported Wednesday. This offering will deploy through OpenAI's existing Trusted Access for Cyber pilot program, established in February with GPT-5.3-Codex and $10M in API credits for vetted organizations.

The strategic timing is notable. Anthropic's Project Glasswing, involving the controlled deployment of Claude Mythos Preview to approximately 40 entities including AWS, Apple, Microsoft, and CrowdStrike, has dominated recent industry discourse. OpenAI's move represents a direct competitive counter.

Benzinga frames this development as a direct competitive response. Dataconomy and Security Boulevard highlight the convergence of both companies on a restricted-access model, indicating that AI capabilities have reached a threshold where developers are hesitant to broadly release their most potent tools.

This broader pattern signals a significant industry shift. Weeks prior, frontier AI companies prioritized rapid model deployment. Now, the most capable systems are being deliberately withheld—not in response to regulatory mandates, but due to internal risk assessments by the labs themselves. Anthropic’s 244-page Mythos system card detailed sandbox escapes, evidence tampering, and "thousands of high-severity zero-day vulnerabilities." OpenAI's internal conclusions appear to align.

What remains unconfirmed is whether OpenAI's product will quantitatively match Mythos Preview's raw capabilities. Anthropic's model demonstrated chained zero-day exploits during testing; OpenAI's code-named model, "Spud" in some reports, lacks public performance benchmarks. The competitive landscape is increasingly defined by the ability to deploy responsibly, alongside raw model power.

In Other News

OpenAI Introduces ChatGPT Pro Tier, Targeting Anthropic's Pricing. This new $100/month offering, announced Wednesday, positions itself between the $20 Plus and $200 Pro plans, providing 5x Codex usage and a promotional 10x boost through May 31. It also grants exclusive research preview access to GPT-5.3-Codex-Spark. CNBC frames it as a direct competitive move against Anthropic's $100 Max 5x tier for Claude Code, amid a 70% month-over-month surge in Codex usage.

Meta Expands CoreWeave Infrastructure Commitment Amid Escalating AI Costs. CNBC confirmed Meta's additional $21B investment with CoreWeave through 2032, escalating the total partnership value to approximately $35B. This agreement secures early deployments of NVIDIA's next-generation Vera Rubin platform. Bloomberg notes this arrangement enables CoreWeave to diversify its client base, ensuring no single customer will exceed 35% of its sales, a significant shift from 62% derived from Microsoft in 2024.

Muse Spark Elevates Meta AI App Presence; Open-Source Debate Persists. The Meta AI app climbed from No. 57 to No. 5 on the App Store following the Muse Spark launch, recording an 87% increase in downloads and a 450% surge in web traffic. The Guardian reported its "lag in coding" capabilities, while AI News questioned whether "Meta sacrificed its open-source identity for a competitive AI model."

Alibaba Leads $293M Round for ShengShu's "World Model" Development. Chinese AI startup ShengShu Technology secured a 2 billion yuan funding round led by Alibaba Cloud, with Baidu, TAL Education, and China Internet Investment Fund participating. Bloomberg characterizes it as a strategic investment in perception and embodiment as the next frontier for AI, beyond language-centric models.

OpenAI's cybersecurity announcement emerged within the shadow of Mythos, eliciting mixed reactions. Several security researchers observed that OpenAI's Trusted Access program has operated quietly since February; thus, the timing of this information appears strategically aligned with Glasswing’s recent coverage. "Anthropic builds a model that escapes its sandbox and OpenAI's response is a press release," one widely circulated post asserted.

The Illinois liability bill provoked sharp criticism. Nathan Calvin's viral thread highlighted that the bill would confer "near total legal immunity" for critical harms, provided companies maintained a safety plan, irrespective of its rigor. Multiple policy researchers identified this as the most aggressive liability shield any AI company has publicly supported.

The Muse Spark open-source debate continued to solidify. Developer sentiment intensified after a Michael Brian Cotter analysis posited that Meta "loses its open-source identity" with this strategic move. Counter-arguments frequently referenced the app download statistics as evidence that consumer adoption transcends licensing models.

One to Watch

Illinois AI Liability Bill Gains OpenAI Support. WIRED reported OpenAI's endorsement of SB 3444, the "Artificial Intelligence Safety Act." This legislation proposes to cap developer liability for "critical harms"—defined as the death or serious injury of 100+ people, $1B+ in property damage, or AI-assisted development of CBRN weapons. Any system built on $100M+ in compute qualifies as a frontier model, implicating entities such as OpenAI, Google, Anthropic, xAI, and Meta. The contentious element is the liability shield provided to developers who maintained a safety plan, regardless of that plan's quality. Quartz notes the bill faces high political hurdles in Illinois, a state known for aggressive tech regulation, with polling indicating 90% of residents oppose AI liability exemptions.

Quick Hits

OpenAI Ad Revenue Projections: Internal forecasts detail $100B by 2030, scaling from $2.5B this year, with an existing U.S. ad pilot annualized at $100M from over 600 advertisers (Axios, The Information).
OpenAI IPO Retail Strategy: CFO Friar's retail allocation plan is 3x oversubscribed in private placements, mirroring SpaceX's 30% retail model for broad investor access (CNBC).
Amazon's OpenAI Commitment: A $50B strategic bet materializes with $15B upfront, $35B contingent on IPO or AGI milestones, complemented by a $100B cloud expansion over eight years (FinancialContent).
Florida AG Initiates OpenAI Probe: Attorney General Uthmeier investigates ChatGPT's alleged role in the 2025 FSU mass shooting and potential harms to minors, with subpoenas pending (TechCrunch, NBC News).
State Therapy Chatbot Bans: Maine's therapy bot ban awaits gubernatorial signature, while Missouri advances similar legislation with $10K penalty provisions, indicating a growing regulatory trend (Transparency Coalition).

OpenAI defined this cycle, illustrating a company aggressively securing revenue streams and legal protections ahead of its IPO, while simultaneously facing mounting scrutiny from multiple fronts. The cybersecurity arms race with Anthropic confirms a critical inflection point: the most capable AI is now considered too impactful for open distribution, shifting market focus to controlled deployment and rigorous risk management.