Case study: Safety standards on California utilities to prevent wildfires
Coby Joseph
October 26, 2023

This case study was written in response to this call for proposals by Holden Karnofsky to inform possible safety standards on leading AI labs. It represents ~65 hours of research, interviews, and writing. I am grateful to Michael Wara (Stanford Law School Senior Research Scholar and consultant for wildfire risk to the California Office of Energy Infrastructure Safety), Melissa Semcer (former Deputy Director of Energy Safety and Principal at Climate Wildfire and Energy Strategies, LLC), Tom Long (Director of Legal Strategy at TURN), and Marc Joseph (former shareholder and President at Adams, Broadwell, Joseph, & Cardozo and my father) to allow me to interview them for this case study.

Background and overview

This case study investigates safety standards applied to investor-owned electrical companies in California (‘utilities’) to reduce the risk of utility equipment igniting catastrophic wildfires. California investor-owned utilities are regulated monopolies responsible for purchasing and distributing electricity across the state. Three utilities cover most of the state, with Pacific Gas and Electric (PG&E) being the largest.1

The field has many possible parallels to the current and possible future landscape for frontier AI labs. Some possible similarities are listed below (though some may become dissimilarities depending on how the AI field evolves).

  • Catastrophic wildfire risk is posed by a small number of companies in California (three utilities cover most of the state).2
  • It is infeasible to fully eliminate catastrophic wildfire risk (at least in short-medium term).
  • The risk of catastrophic wildfires has quickly increased in the past two decades.
  • During this time, risk levels and the individual sources of risk have been highly uncertain (though they are better understood now).
  • The cost and effectiveness of risk mitigation strategies has also been largely unclear.
  • Safety standards partially resemble a licensing scheme, with approval based on a fairly subjective evaluation across a wide range of metrics.
  • Safety measures can accompany economic tradeoffs for the general public (namely higher electricity rates).
  • Reputational risks play an important factor for utilities, as the largest utilities are well known by the general public and the risk posed by such companies is widely discussed in California media and politics.

This case study aims to highlight examples of company actions and regulator successes and challenges in this (potentially) similar environment, to help calibrate our expectations for AI company behavior and the risks and opportunities for safety standard strategies.

Based on my research, the following features of safety standard development on California utilities to prevent catastrophic wildfires should provide (further) caution for AI safety standard setting:

This case study also explores possible benefits and challenges of several regulatory strategies that could be relevant for the AI field.

Finally, the case study highlights:

Contents

This case study is structured as follows:

Bullet points with arrows next to them can be clicked to reveal additional detail.

The need for safety standards

California wildfires cause significant death and destruction.
The risk of wildfires has significantly increased in recent decades, almost certainly due to climate change-induced droughts.
Equipment from California’s utilities is a significant source of wildfires that have killed hundreds of people and caused billions of dollars in damage.
There is no simple and affordable solution to eliminate wildfire risk from utility equipment.

Current standards

  • The California Public Utilities Commission (CPUC) is the primary regulator of utilities. In 2022, the Office of Energy Infrastructure Safety (‘Energy Safety’) was created outside of the CPUC to oversee utility processes related to reducing wildfire risk.
Energy Safety issues “safety certifications'' to qualifying utilities. Failure to receive a safety certification could significantly harm a utility’s financial standing.
  • Core requirements for a safety certification include:
    a safety culture assessment,
    • an approved executive compensation structure that ties compensation to safety performance. This requirement is explained in more detail here.
    • an annually approved Wildfire Mitigation Plan (WMP). WMPs “describe how the electrical corporation is constructing, maintaining, and operating its electrical lines and equipment in a manner that will minimize the risk of catastrophic wildfire.” WMP evaluation also includes an assessment for a utility’s capacity to execute the plan. The text box below provides more details about the WMP components.
    • Demonstration of implementing its approved WMP, and board-of-director-level safety reporting structure.
    Auditing includes field inspections, independent evaluations, and annual assessments of compliance with the WMP. Energy Safety can recommend to the CPUC to impose penalties on utilities “for noncompliance with its approved plan.”
    Wildfire Mitigation Plan components

    Key takeaways

    Proactivity vs. reactivity

    • When starting from a low base, creating and enforcing standards can be a lengthy process – it took roughly 5-10+ years after it was clear that catastrophic risk was present and rising for standards to be created and have a meaningful impact on all utilities. This point is summarized in the bullets below and explored in more detail in this text box.
      • Until legislation required stricter regulation (particularly starting in 2017), regulators lacked sufficient capacity and will to promptly impose rules in the face of opposition from most utilities.
      • Following legislation, it took several years to build internal regulatory structure and expertise.
      • For the first year or two under new regulatory structure, regulators approved safety plans with serious deficiencies on the condition that utilities would build capacity for improving safety over time.
    Amid a changing risk landscape, utilities have appeared not to take tail risk sufficiently seriously until they experience it first hand, even when such risk would threaten company profits. More concerningly, regulators have generally not mandated substantially stricter safety standards for utilities until that utility has caused a disaster. Some events that look like ‘warning shots’ in hindsight did not sufficiently spur urgent safety improvements.
    The evolution of safety standards

    Utility motives

    • On its surface, the California utility market appears more likely than most capitalist markets to be one in which companies prioritize public welfare, including for the following reasons:
      • Utilities have a highly regulated business model.
      • Wildfire risk has become clear after utilities have caused deadly catastrophes.
      • Utilities face significant reputational challenges. PG&E in particular has received significant negative publicity for many years related to wildfires. Its employees live among their customers, and according to one expert, employees have faced significant safety concerns while in the field due to apparent threats from members of the general public.
    Still, while experts believe public welfare and reputational concerns have played a role in utility decisions, most decisions (especially by PG&E) would be best predicted by utilities’ profit maximizing incentive structure, including consistent underspending on maintenance.
    Experts indicated that legal liability – established by courts rather than the legislature – has proved substantially more influential on company behavior than public penalties.
    Tying executive pay to safety outcomes may meaningfully influence utility behavior, but the mechanism by which this is done may make an important difference.

    Activists and third parties

    Even when activists concerned about tail risks had little early success (from what I can tell), their proposals appeared to set the foundation for later standards.
    The CPUC intervenor compensation program – which pays individuals and organizations when their input has been useful to a CPUC proceeding – could be a replicable model to build an ecosystem of third party experts.

      Regulatory structure, processes, and challenges

      Energy Safety was deliberately created to be a separate regulatory body from the CPUC to focus on regulating wildfire risk by California utilities. My impression from conversations with experts is that having a narrowly safety-focused organization has helped build significant regulatory expertise and make progress establishing a robust oversight process. One expert also suggested the system creates a healthy tension between Energy Safety’s focus on efficient risk reduction and the CPUC’s focus on spending efficiency.
      Still, separating the primary safety regulator from the agency that controls funding for safety measures has led to several challenges, too. Experts indicated that utilities may have insufficient clarity for how to respond to conflicting directions between the safety regulator and the agency that controls funding.
      One expert indicated to me that the cost and effectiveness of safety-enhancing strategies has been highly uncertain (though perhaps less so today than it used to be). The separate processes and different timelines between Energy Safety and the CPUC appear to risk slowing safety investments and reduce Energy Safety's flexibility to quickly update guidance as the cost and effectiveness of various safety strategies becomes clearer.
      Lengthy auditing and investigation periods that reduce regulators’ ability to update oversight strategies appear to be particularly costly amid these uncertainties. For example, Energy Safety appears to have incomplete information about utilities’ compliance with the prior year’s WMP when evaluating its subsequent WMP.
      Regulation is also challenged by the state’s dependence on utilities. This case study highlights the risk of using a licensing scheme to regulate companies that become ‘too-important-to-fail.’ Failing to issue safety certifications to utilities that insufficiently mitigate catastrophic risk would threaten utilities' financial standing and could lead to higher electricity prices.11 This may make regulators reluctant to deny certification, thereby reducing pressure on utilities to fully comply with safety standards. (A similar challenge could emerge if governments and businesses become dependent on (updates to) AI systems).

      Influence of voluntary safety actions

      Early voluntary safety practices by San Diego Gas and Electric (SDG&E) appear to have influenced subsequent safety standards applied to all large utilities. SDG&E upgraded fire prevention efforts after it was sued by residents for causing a devastating wildfire in 2007. Since then, it undertook a wide range of voluntary safety practices, and was significantly more supportive of regulation attempts than other utilities, especially PG&E.
      Regulators cite peer utilities’ safety practices when evaluating WMPs. This competitive pressure appears to embed (sometimes voluntary) best practices into safety standards.

      Risk assessment

      For many years, regulators and utilities appear to have had insufficient urgency to develop a robust and detailed risk identification and prioritization methodology, though risks are dramatically better understood and modeled today.
      Still, some experts argue that risk models may focus too heavily on median wildfire risk rather than potential catastrophic fires.

      Audits and penalties

      After causing a devastating wildfire, utilities generally face penalties from the CPUC and courts (if they were found to be at fault), in addition to being liable for damages caused to victims. Penalties from courts have generally been small relative to utility revenues. Some CPUC penalties have been much larger, though they are still much smaller than liabilities facing utilities after causing a wildfire.
      The CPUC does not appear to substantially use audits and penalties to punish utilities for failing to comply with its WMP.

      If you have any comments or questions about this case study, feel free to email me at coby.joseph@gmail.com.

      1: This case study focuses on safety standards on the three largest utilities and may not accurately represent standards applying to smaller utilities.
      2: This may or may not be a parallel to the AI field in the future.
      3: Katherine Blunt, California Burning, 2022. Page 268.
      4: ibid. Page 230.
      5: ibid. Page 211.
      6: ibid. Page 211.
      7: ibid. Page 100.
      8: Available in the ‘Approved 2023 Executive Compensation Guidelines – 11/28/22’ accessible here
      9: Katherine Blunt, California Burning, 2022. Page 188.
      10: Available here, see Energy Safety’s 2020 Annual Reports on Compliance: PG&E – 2/24/23
      11: As it would increase the utility’s risk of being unable to recover money from the California Wildfire Fund if it faced significant liability, likely raising its borrowing costs and increasing the risk of future bankruptcy, as indicated to me by experts.
      12: “PG&E 2023-25 Wildfire Mitigation Plan Revision Notice – 6/22/23” available here
      13: Katherine Blunt, California Burning, 2022. Page 151.