SIGNAL

Anthropic has reached $30B in annualized revenue run rate, positioning it as a credible alternative to OpenAI ahead of OpenAI's anticipated IPO process. The company is previewing two initiatives—Project GlassWing and Claude Mythos—with the latter representing the first model Anthropic has deemed too risky to publicly release since GPT-2's era, suggesting either significant capability gains or safety concerns that warrant containment. This timing and messaging strategy directly counter OpenAI's market narrative by demonstrating both commercial traction and responsible restraint in capability deployment. The move signals Anthropic is competing on both business fundamentals and on the increasingly important dimension of trustworthiness in AI governance.

OpenAI has deployed what it calls a "Dark Factory"—a fully automated code generation and deployment system handling 1 million lines of code daily with zero human authorship or review, according to researcher Ryan Lopopolo. The system processes approximately 1 billion tokens per day, operating autonomously across OpenAI's Frontier and Symphony projects without manual code inspection checkpoints. This represents a fundamental shift in how large AI labs approach infrastructure scaling: moving from human-reviewed development to autonomous code generation at production scale. The implications are significant for software reliability, security governance, and competitive dynamics, as it demonstrates AI systems can now handle complex engineering tasks independently—though the operational risks of zero human oversight remain undisclosed.

Iranian state-sponsored hacking groups have successfully disrupted operational systems at multiple US critical infrastructure facilities, marking an escalation in cyber activity correlated with heightened US-Israel military operations in the Middle East. The attacks targeted industrial control systems rather than just data theft, indicating capability to cause physical disruption to power, water, or manufacturing networks. This represents a shift from espionage to potential destructive intent and demonstrates Iran's willingness to move beyond rhetoric to direct retaliation against US infrastructure. The incidents underscore a growing asymmetric threat vector for US national security as regional tensions rise.

Russia's military cyber unit (GRU) compromised thousands of consumer routers across 120 countries in a widespread campaign targeting home and small office networks. The attacks focused on end-of-life router models—devices no longer receiving security patches—making them easy entry points for persistent access. This matters because compromised routers give attackers an invisible foothold in networks, enabling espionage, lateral movement to connected devices, and potential staging for operations against higher-value targets while remaining difficult to detect. The scale and targeting of non-enterprise infrastructure suggests preparation for broader operations rather than immediate financial gain.