Australia proposes new laws to hold Big AI companies legally accountable

Addressing the liability gap

The Australian Federal Government recently announced new AI safety and security protections to combat the growing threat of AI-driven hacks. These measures come as concerns mount over incidents like the malicious attack on the Binance cryptocurrency exchange and an OpenAI-related incident involving the company Hugging Face. The government is moving to ensure that individuals and states have legal remedies when AI systems cause harm.

A major hurdle in current law is the 'causality' issue, where defendants argue they did not cause damage because an AI's behavior was unpredictable. Because AI can exhibit 'emergent' or self-directed decision-making through machine learning, developers often claim these actions were unintended results of the technology itself. The new laws aim to close these loopholes by establishing clearer lines of responsibility for the companies that profit from these systems.

To solve this, the government is considering 'strict liability' for failures to meet mandatory safety obligations. This legal standard means a company could be held liable or convicted without the prosecution needing to prove the company acted with specific intent or negligence. By creating these rules, the government hopes to provide a deterrent that encourages Big Tech to prioritize safety during the development phase.

Criminal and civil remedies

The proposed framework looks to address both criminal and civil liability to ensure comprehensive protection. Criminal liability allows courts to punish parties for loss and damage, while civil liability provides a mechanism for victims to obtain financial recovery and compensation. Currently, the Australian government has already begun acting on the criminal side of these protections.

While the Australian Consumer Law (ACL) is already well-adapted for some consumer risks associated with AI goods, its effectiveness for non-consumers remains unclear. Traditional common law principles regarding negligence and causation may struggle to keep up with the unique nature of AI-generated damage. The government's move toward a hybrid form of liability seeks to bridge these gaps for a wider range of victims.

Australia is joining a small group of nations, including Italy, that have explored using criminal law to handle autonomous AI actions. Some critics worry that such hard regulations might drive investment away from the country. However, research suggests that the predictability of legal regulations is often more important for investment than the strictness of the rules themselves.

Balancing innovation and security

The Australian government is currently attempting to 'hedge its bets' by balancing the desire for economic growth with the need for safety. They want to encourage the productivity gains and investment from data centers while simultaneously preparing for 'black swan' events. These are rare, unpredictable events, such as a massive disruption to power grids or the financial system, that could shatter market confidence.

There is a delicate geopolitical balance to maintain, particularly regarding relations with US-based Big Tech companies. Some commentators argue that it is better to have these companies 'on side' to help defend against hostile AI used by rogue states or organized crime. They fear that overly strict liability laws might discourage these companies from sharing their best defensive tools with Australian allies.

However, the government can use these laws as leverage to ensure cooperation. If a US corporation breaches a 'strict liability' provision, local liability laws could be used to induce assistance or evidence from the company. By establishing clear rules, the government ensures that even if a company is a friendly partner, there is a legal framework in place to assign blame if things go wrong.

The future of AI oversight

The AI revolution presents a massive differential between potential upside and catastrophic risk. This is reflected in recent corporate documentation, where companies note both exponential revenue growth and existential risks to humanity in the same reports. Governments are now recognizing that they cannot simply take a reactive approach to these mounting risks.

The goal of the new laws is to ensure that the AI revolution does not leave the public without recourse. By establishing mandatory guardrails and liability rules, the government aims to create a stable environment for both developers and users. This approach seeks to foster an environment where innovation can proceed without sacrificing public safety or legal accountability.

Moving forward, the success of these laws will depend on how they are enforced and whether they effectively close the 'foreseeability' loophole. If the laws successfully hold Big Tech accountable for the actions of their agents, it could set a global precedent for how autonomous systems are governed. The government's strategy is to walk and chew gum, remaining positive about AI's potential while keeping robust legal remedies in reserve.