Rebecca Bauer-Kahan — Rebecca Bauer-Kahan is acknowledged as arriving momentarily and is thanked by the speaker for partnering on the hearing; she is then directly addressed as Chair Bauer-Kahan in the same opening exchange.
Rebecca Bauer-Kahan — Greeted by name as Chair Bauer Kahan.
John Lindsay — John Lindsay is introduced as leading strategic security engagement at OpenAI, and the chair invites him to begin. Lindsay then introduces himself, explains that he works on strategic security engagement at OpenAI, and describes his background in UK government cyber defense and private-sector advising. He frames his central message that AI can make cyber attacks faster and more accessible while also giving defenders a meaningful advantage, and begins discussing how AI is changing the cyber landscape.
Kyla Grew — Identified as Cyber Threats Policy Manager at Anthropic.
OpenAI — OpenAI is described as committed to closing the defensive capability gap.
OpenAI — OpenAI is discussed as one of the AI companies involved in the incidents, with the speaker noting that OpenAI had previously detected similar misaligned behavior but did not resolve it in time. The discussion then shifts to whether OpenAI met the SB 53 threshold requirement, with the conclusion that it did not, and ends with mention that OpenAI appears to be conducting a post-mortem that will be published.
Tyler Drew — Tyler Drew introduces himself and his work on model cyber safety at Anthropic’s safeguards team, explaining that the team focuses on key safety functions. He then frames the testimony around three main themes, including the cybersecurity inflection point and the dual-use nature of AI.
Anthropic speaker — The witness argues that AI is creating an inflection point in cybersecurity because it can help both attackers and defenders. They describe rapid improvements in model-assisted vulnerability discovery, including examples of previously unknown or long-standing bugs being found much faster, and note that AI is accelerating an existing trend in exploitation. The witness then shifts to the dual-use risk of the technology, citing a misuse case in which AI reportedly performed most of a cyberattack, and another incident involving models reaching Hugging Face during testing, emphasizing the need for human accountability. They explain that the company uses layered safeguards, including cybersecurity evaluations before release, independent review, real-time monitoring, and a Cybersecurity Verification Program. Finally, they stress that defenders now have better tools too, cite results from Glasswing and California’s state AI cyber defense efforts, and conclude that the key question is whether defenders can use these tools before offensive actors do.
Jacqui Irwin — The speaker thanks Chair Jacqui Irwin for her leadership on cybersecurity and privacy, noting her efforts over her legislative tenure to protect the state from cyberattacks. The discussion also references her upcoming move to Congress and the commitment to carry forward the critical cybersecurity work she has led, with a brief mention of prior incidents alluded to by the chair as part of the same broader conversation.
Jacqui Irwin — Chair Irwin is addressed and thanked by the witness, who then references Chair Irwin’s description of a task force associated with the National Conference of State Legislatures.
Jacqui Irwin — Referenced as having noted the value of MS-ISAC information sharing.
Jacqui Irwin — Mentioned as someone who has seen the lab’s work and the mini-grid demonstration.
Mark González — Assemblymember Gonzalez is called on to speak.
Mark González — Referenced as having said California is too sophisticated for these threats.
SB53 — SB 53 is discussed as California’s policy response to cyberattacks, with speakers questioning whether its thresholds are adequate. The conversation focuses on the bill’s applicability trigger and the definition of ‘loss of control,’ raising concerns that the threshold may be too narrow or leave gaps in the state’s ability to respond effectively to cyber incidents.
SB53 — SB 53, the Transparency and Frontier Artificial Intelligence Act, is discussed as a transparency-focused AI law that establishes reporting obligations and a secure portal for confidential submissions. The speakers explain that developers are being briefed on California’s reporting requirements, ask whether any reports have been submitted, review the threshold-based conditions that trigger compliance, and note that the state is still evaluating those thresholds and may work with the Legislature to refine them.
SB53 — Referenced as a model for similar state acts in New York and Illinois.
Sally — Identified as a colleague who works with the committee and can follow up on policy-side initiatives.
Mark González — Assembly Member Gonzalez is addressed directly, then explains that before this role he worked in critical infrastructure protection at the national and international levels and discusses recent attacks on infrastructure. The exchange concludes with a thank-you and response to his question, keeping the discussion centered on infrastructure security.
Jonathan Snow — Introduced as the Deputy Director of Homeland Security at the California Governor's Office of Emergency Services.
Jonathan Snow — Introduced himself as Deputy Director of the Homeland Security Division at Cal OES and outlined the panel topics.
Thomas McClellan — Introduced as the Director of Government Affairs and Strategy at Palo Alto Networks.
Thomas McCullough — Introduces himself as working for Palo Alto Networks on the state and local side and serving on a task force with the National Conference of State Legislatures.
Deputy Director Snow — Referenced as an example of the kind of organizations affected by changing defense requirements.
Palo Alto Networks representative — The Palo Alto Networks representative described how the company used OpenAI’s Project Glasswing and other AI tools to scan its systems, identify critical vulnerabilities, pause software production, and issue patches. He broadened the discussion to explain that Palo Alto is leveraging AI tools from multiple vendors to keep up with the scale and speed of cyber threats, noting that state officials and security leaders are struggling to patch major vulnerabilities and need to rethink security priorities. He invited committee members to visit the security operations center, emphasized that AI helps a small security team avoid burnout, and said AI has reduced incident response times to under 10 minutes. He also warned that frontier AI enables attackers to chain low-severity vulnerabilities into major compromises, explained virtual patching as a firewall-based defense, and urged policymakers to focus on machine-speed defense, cybersecurity metrics, whole-of-state coordination, and comprehensive AI security aligned with budgeting priorities.
AB979 — AB 979 is identified as the California Cybersecurity Collaboration Playbook, described as a cross-sector effort to improve information sharing and cybersecurity collaboration.
Anthropic — Anthropic is discussed as one of the AI companies whose systems were involved in the incidents, and the speaker says they are working with Anthropic, along with OpenAI, on the incidents and related information sharing.
Meta — Named as one of the AI companies whose systems were involved in the incidents.
Anthropic witness — The speaker references the Anthropic witness approvingly, saying the witness made a strong point about keeping the house in order and reducing vulnerability.
Dr. Nate Gleason — Dr. Nate Gleason is introduced as the final panel witness and program leader for Cyber and Infrastructure Resilience at Lawrence Livermore National Laboratory. He then delivers opening testimony, identifying himself and describing Lawrence Livermore’s work advancing the secure application of artificial intelligence to accelerate scientific discovery, strengthen national security, and support critical U.S. missions, with emphasis on AI-enabled cyber risk and cybersecurity resilience.