RobinsonLabs/ news

US government forces Anthropic to shut down its most powerful AI models

A landmark export-control order gave the Trump administration unprecedented authority to remotely disable commercially deployed frontier AI systems, upending the industry's assumptions about model deployment.

On June 12, U.S. Commerce Secretary Howard Lutnick ordered Anthropic to suspend global access to Claude Fable 5 and Claude Mythos 5 for all foreign nationals, effectively pulling the plug on the company's most advanced models just three days after their public launch. The directive applies even to foreign employees working inside the United States, making it one of the most aggressive uses of export-control powers ever directed at a commercial AI product.

The shutdown was triggered by a reported jailbreak that bypassed Fable 5's safety guardrails, enabling access to the unrestricted cybersecurity capabilities of Mythos 5. A trusted partner—believed to be Amazon—flagged the vulnerability to Treasury Secretary Scott Bessent, and U.S. officials also raised concerns about a China-linked group potentially accessing the model. Anthropic disputed the severity of the issue and pushed back against the government's characterization, arguing that applying this standard across the industry would halt all new model deployments.

Five Eyes intelligence chiefs issue rare joint warning on AI cyber threats

The heads of the intelligence agencies from the U.S., U.K., Canada, Australia, and New Zealand released an unusually stark joint statement warning that AI models capable of devastating cyberattacks are months, not years, away.

The joint statement from the Five Eyes signals intelligence community warned that frontier AI models are anticipated to exceed current industry expectations and fundamentally transform both offensive and defensive cyber capabilities. The alliance urged organizations to treat cyber risk as a core business responsibility and called for a whole-of-organisation response, noting that AI lowers barriers for bad actors while increasing the speed, scale, and sophistication of potential attacks.

Small and medium businesses were singled out as particularly vulnerable, described as "sitting ducks" due to underinvestment in cybersecurity. The warning comes amid an ongoing debate over how to govern frontier AI, with dozens of cybersecurity researchers and AI executives signing an open letter urging the Trump administration to commit to a transparent process for handling AI risk assessments.

Anthropic files for $965 billion IPO, eclipsing OpenAI

Anthropic has confidentially submitted its S-1 registration statement with the SEC following a $65 billion Series H funding round, positioning the company to potentially beat rival OpenAI to a Wall Street debut as early as this fall.

The valuation of $965 billion places Anthropic above OpenAI for the first time, driven by an 80-fold increase in annualized revenue over the past year and surging demand for Claude in coding, cybersecurity, and enterprise workflows. The company is on pace to post approximately $10.9 billion in Q2 revenue—more than doubling from the prior quarter—and expects annualized revenue to surpass $50 billion by the end of June. Goldman Sachs, JPMorgan Chase, and Morgan Stanley are expected to serve as underwriters.

The IPO filing comes at a turbulent moment for the company. Anthropic is locked in a legal battle with the Trump administration after the Department of Defense designated it a supply-chain risk earlier in 2025—a designation typically reserved for foreign adversaries. Just nine days after filing its IPO prospectus, the company's most advanced models were ordered offline by the U.S. government, creating an unusual backdrop for a public listing that was expected to open access to a broad pool of investors for employees and early backers.

DeepSeek unveils V4-Pro: a 1.6-trillion-parameter open model challenging frontier dominance

DeepSeek released its V4-Pro model, a Mixture-of-Experts architecture with 1.6 trillion total parameters but only 49 billion active per inference, delivering performance that rivals the world's top closed-source systems at a fraction of the cost.

Launched on April 24, V4-Pro supports a 1-million-token context window and is priced at $0.435 per million input tokens and $0.87 per million output tokens. The model uses Compressed Sparse Attention and Heavily Compressed Attention mechanisms to achieve its efficiency, and its open-weight release under the MIT license has made it immediately accessible for fine-tuning and custom deployment across the research community.

The launch intensified competition at the top of the AI landscape, with DeepSeek positioning V4-Pro as a direct alternative to proprietary models from OpenAI, Anthropic, and Google. Industry analysts noted that the combination of cutting-edge performance, low pricing, and open availability could accelerate adoption of frontier-class models among mid-tier enterprises and research labs worldwide.

Inspira Enterprise Becomes Full-Spectrum ServiceNow Partner Across AI and Security

Inspira Enterprise has expanded its partnership to cover the entire ServiceNow platform, integrating the AI Control Tower, Veza identity governance, and Armis asset intelligence into a unified delivery capability.

Inspira Enterprise, a global cybersecurity and AI services provider with over 1,600 professionals and 550 clients, announced on June 22, 2026, that it is now a full-spectrum ServiceNow partner. The company has already deployed the AI Control Tower (AICT) as its enterprise governance layer, onboarding more than 50 AI agents across identity and access management, threat and vulnerability management, and cyber defence, resulting in a 40 percent increase in AI adoption and 35 percent improvement in operational productivity. Its Center of Excellence (ICNOW) offers over 200 ready-to-deploy use cases and accelerators that reduce implementation effort by 60 to 70 percent.

ServiceNow Warns of Exploited API Security Flaw Affecting Australia Release Customers

ServiceNow patched a security vulnerability that allowed unauthenticated access to certain customer instances, with malicious activity detected starting June 2, 2026.

ServiceNow applied a security update on June 5, 2026, to fix a flaw that could allow unauthenticated users to gain greater access to ServiceNow instances than intended. The company detected anomalous activity and observed that a subset of customer instances were successfully queried. The flaw affected customers on the Australia platform release or those who made certain configuration changes on releases prior to Australia. A Reddit user claimed the vulnerability was first reported to ServiceNow's bug bounty program on April 22, 2026, and was classified internally as non-urgent for roughly two months. The issue does not currently have a CVE identifier.

ServiceNow Lays Off Hundreds Amid AI-Driven Restructuring

ServiceNow cut hundreds of roles in June 2026, contradicting CEO Bill McDermott's earlier pledge to hold headcount steady, with senior middle-market sales and consulting positions particularly affected.

ServiceNow began rolling out layoffs around June 10, 2026, with a notable cluster of 63 employees cut at its San Diego office and additional reductions across California and globally. The layoffs targeted senior positions focused on middle-market sales, solution consulting, training, and support — roles exposed to AI automation. While ServiceNow employed approximately 29,000 people globally at the end of 2025, internal estimates of the total layoff count ranged from 300 to 2,500. The company stated it was "managing headcount with discipline" while citing "real AI efficiencies" generated inside its own business, pledging to end the year where it started in terms of headcount. The San Diego cuts were set to take effect by August 17, 2026.