President Donald Trump signed an executive order on June 2, 2026, titled “Promoting Advanced Artificial Intelligence Innovation and Security.” The order directs federal agencies to collaborate with the private sector to modernize systems and protect them against external threats. A primary component involves establishing a voluntary system for government evaluation of advanced frontier AI models before public release to assess cybersecurity risks. The National Security Agency (NSA) is tasked with developing a classified benchmarking process to identify if models could be exploited as powerful tools for cyberattacks. White House advisor David Sacks noted that the framework allows AI labs to comply without delaying releases, citing a reduction in the required evaluation period from 90 days to 30 days. Meanwhile, OpenAI has proposed an alternative regulatory approach, advocating for mandatory risk evaluations overseen by civilian agencies rather than the current government-led structure.

Anthropic has announced an expansion of Project Glasswing, an initiative dedicated to securing software through vulnerability scanning. Since its launch in early April 2026, roughly 50 initial partners utilizing the Claude Mythos Preview model have identified over 10,000 high- or critical-severity security flaws in their codebases. The project is now extending to approximately 150 new organizations, provided they meet specific security requirements. Anthropic’s internal research on AI-enabled cyber threats indicates that malicious actors are increasingly using AI for complex, autonomous operations, rendering traditional risk-differentiation methods less effective. Goldman Sachs CEO David Solomon confirmed that the firm is among those testing the Mythos model, emphasizing the need for broad information sharing and security guardrails.

Separately, Perplexity CEO Aravind Srinivas has identified “token value per watt” as the critical metric that will determine the long-term winners in the AI industry. This objective balances accuracy, latency, cost, privacy, and intelligence to measure energy efficiency against economic output. Perplexity’s perspective highlights the growing importance of an orchestration layer to manage the energy demands of agentic AI (systems capable of executing complex tasks autonomously). Experts suggest that maximizing this ratio will be central to how companies navigate the rapid progression of AI capabilities through 2028 as organizations work to adapt to the pace of technological shifts.

President Donald Trump signed an executive order on June 2, 2026, titled “Promoting Advanced Artificial Intelligence Innovation and Security.” The order directs federal agencies to collaborate with the private sector to modernize systems and protect them against external threats. A primary component involves establishing a voluntary system for government evaluation of advanced frontier AI models before public release to assess cybersecurity risks. The National Security Agency (NSA) is tasked with developing a classified benchmarking process to identify if models could be exploited as powerful tools for cyberattacks. White House advisor David Sacks noted that the framework allows AI labs to comply without delaying releases, citing a reduction in the required evaluation period from 90 days to 30 days. Meanwhile, OpenAI has proposed an alternative regulatory approach, advocating for mandatory risk evaluations overseen by civilian agencies rather than the current government-led structure.

Anthropic has announced an expansion of Project Glasswing, an initiative dedicated to securing software through vulnerability scanning. Since its launch in early April 2026, roughly 50 initial partners utilizing the Claude Mythos Preview model have identified over 10,000 high- or critical-severity security flaws in their codebases. The project is now extending to approximately 150 new organizations, provided they meet specific security requirements. Anthropic’s internal research on AI-enabled cyber threats indicates that malicious actors are increasingly using AI for complex, autonomous operations, rendering traditional risk-differentiation methods less effective. Goldman Sachs CEO David Solomon confirmed that the firm is among those testing the Mythos model, emphasizing the need for broad information sharing and security guardrails.

Separately, Perplexity CEO Aravind Srinivas has identified “token value per watt” as the critical metric that will determine the long-term winners in the AI industry. This objective balances accuracy, latency, cost, privacy, and intelligence to measure energy efficiency against economic output. Perplexity’s perspective highlights the growing importance of an orchestration layer to manage the energy demands of agentic AI (systems capable of executing complex tasks autonomously). Experts suggest that maximizing this ratio will be central to how companies navigate the rapid progression of AI capabilities through 2028 as organizations work to adapt to the pace of technological shifts.