Anthropic Announces Mythos AI with Potential Grid Security Applications

recently announced its Mythos AI model, which can autonomously detect software vulnerabilities at a large scale and high speed. com reported that this capability applies to power grid systems running on aging software, potentially serving as both a threat and a tool for improvement.

The U.S. power grid consists of more than 3,000 utilities operating 600,000 miles of interconnected power lines, supporting all 16 critical infrastructure sectors as defined by the U.S. Cybersecurity and Infrastructure Security Agency (CISA). Grid operators have used digital software since the 1960s, including Supervisory Control and Data Acquisition (SCADA) systems to monitor and manage power flows.

Utilities have added new software layers over time, resulting in varied and often outdated systems across substations and energy management platforms. These tools provide operators with visibility into power flows, overloads, and rerouting needs.

The 2003 Northeast Blackout illustrated the consequences of software failure, where a failed alarm left operators unaware of an overloading power line. This led to outages affecting 50 million people across eight U.S. states and Canada, causing an estimated $6 billion in damage and contributing to 11 deaths.

com noted that the incident turned a containable problem into the largest blackout in North American history due to loss of system visibility. In April 2026, CISA issued an advisory stating that hackers had exploited SCADA systems, resulting in operational disruptions and financial losses.

Additionally, in 2024, a China-linked group called Volt Typhoon infiltrated a Massachusetts electric utility and remained undetected for 10 months. Security firm Dragos, which investigated the breach, reported that the intruders copied network diagrams and collected sensor data without triggering alarms or causing outages.

The grid's vast scale, old code, and fragmented ownership make comprehensive security challenging. Anthropic has launched Project Glasswing, providing about 50 vetted organizations access to Mythos to identify and patch vulnerabilities before exploitation.

Energy is among the sectors that could benefit from this initiative. Coordinated remediation across utilities remains complex due to diverse governance and budget structures. com stated that applying such AI capabilities broadly to the grid represents a shift in addressing security needs.