AI Safety
Jul 11, 2026
Why Frontier AI Still Hallucinates, and What to Do
Even the most advanced AI models continue to fabricate facts, citations, and events with alarming confidence. This deep-dive explores recent hallucination failures and unpacks the architectural reasons they're so hard to eliminate.
Hallucination
Large Language Models
AI Reliability
Jul 10, 2026
Inside the Black Box: How the US Cleared OpenAI's Model
The U.S. government signed off on OpenAI's latest frontier model — but the details of that safety review remain murky. What criteria were used, who was involved, and what does this mean for AI oversight going forward?
OpenAI
AI Safety
Government Oversight
Jul 7, 2026
Anthropic's J-Lens Finds a Hidden 'Workspace' in Claude
Anthropic researchers have discovered that Claude spontaneously developed an internal structure mirroring global workspace theory — a leading neuroscientific model of human consciousness. The finding, built around a new interpretability tool called the J-lens, has already begun reshaping how Anthropic monitors Claude for safety risks, including detecting silent strategic reasoning that never surfaces in the model's output.
Anthropic
Claude
AI Safety
Jul 1, 2026
Anthropic Strikes Deal to Restore Claude Fable 5 Access
Anthropic reached an agreement with the Trump administration to extend a key safety guardrail on its Claude Fable 5 model, prompting the Commerce Department to lift export controls that had effectively taken the model offline. The fix targets a specific jailbreak behavior flagged in an Amazon research paper — but the company's troubles with the Pentagon aren't fully resolved.
Anthropic
AI Safety
AI Policy
Jul 1, 2026
Trump Admin Lifts Export Curbs on Anthropic's AI Models
The Commerce Department is removing export restrictions on Anthropic's most powerful AI models after the company struck a deal to strengthen safety safeguards. The move follows weeks of behind-the-scenes negotiations — and a notable shift in how Anthropic engaged with the administration.
Anthropic
AI regulation
export controls
Jun 26, 2026
White House Pressures OpenAI to Delay GPT-5.6
The Trump administration has reportedly asked OpenAI to hold back its latest model, GPT-5.6, from a broad public launch. Instead, OpenAI plans to share the model with a limited group of select partners while it navigates the White House's safety concerns.
OpenAI
GPT-5
AI Safety
Jun 26, 2026
Patronus AI Raises $50M for Simulated 'Digital Worlds'
Patronus AI, founded by former Meta AI researchers, has closed a $50M funding round to develop simulated environments that rigorously stress-test AI agents. Investor demand signals the market's growing urgency around AI reliability and safety infrastructure.
AI Agents
Funding
AI Safety