Wired, March 29, 2023
As artificial intelligence (AI) becomes more prevalent in society, the issue of its security becomes increasingly important. Red teaming, the process of attempting to get an AI system to act in harmful or unintended ways, is a valuable step towards building AI models that won’t harm society. External experts and public reports can be used to identify potential harms and mitigate them before they cause significant problems in the real world. However, red teaming alone is not enough to overcome the harms uncovered. Violet teaming, the process of identifying how a system might harm an institution or public good, and then supporting the development of tools using that same system to defend the institution or public good, is needed to redirect the power unleashed by AI systems to defend public goods.
To execute violet teaming, a “resilience incubator” can be used, which pairs grounded experts in institutions and public goods with people and organizations who can quickly develop new products using the AI models to help mitigate the risks. For example, developing contextualization engines that can reduce the harms of hyper-targeted scams and disinformation by helping people navigate a rapidly evolving information environment. While AI companies sometimes provide early access or economic support to product developers, that is primarily for profit and not to ensure societal resilience in the face of broader access. Therefore, there are currently few incentives to do red teaming or violet teaming, let alone slow down AI releases enough to have sufficient time for this work.
Governments need to act to incentivize companies to prioritize the security of AI systems. Independent governance processes at a national or even global scale can make critical decisions, such as what kinds of testing and guardrails are necessary for model release, more democratically. Such processes can be facilitated by a neutral third party, with extensive access to diverse experts and stakeholders. Ultimately, to get through the next phase of the AI revolution intact, we need to proactively mitigate risks in the systems themselves through red teaming, protect against their impact through violet teaming, and decide what guardrails we need around such use cases through democratic innovation.
UK Gov, March 29, 2023
The UK government has launched a new white paper to provide guidance on the use of artificial intelligence (AI) in the country. The document outlines a new approach to regulating AI that will build public trust and make it easier for businesses to innovate, grow and create jobs. The aim is to unleash the benefits of AI, which already contributes £3.7 billion to the UK economy, and help the technology develop in a safe and responsible way.
To achieve this, the white paper proposes five principles, including safety, transparency, and fairness, which will guide the use of AI in the UK. The document outlines the need for accountability and governance, measures to ensure that AI is functioning securely, and ways to make sure that people can dispute harmful outcomes or decisions generated by AI. The government will avoid heavy-handed legislation that could stifle innovation and will empower existing regulators to come up with context-specific approaches to AI governance.
Over the next 12 months, regulators will issue practical guidance and tools to help organizations implement the principles of the white paper in their sectors. The government will also fund a new sandbox trial to help businesses test AI rules before getting to market. Organizations and individuals working with AI can share their views on the white paper as part of a new consultation process. By implementing the principles outlined in the white paper, the UK will ensure that the rules can adapt to the rapid development of this technology and that it can be used to deliver stronger economic growth and better jobs, while radically improving people’s lives.
VentureBeat, March 28, 2023
Artificial intelligence (AI) has become a powerful tool for both good and bad actors. ChatGPT, a popular AI chatbot, has been used by cybercriminals to create malware, ransomware, and phishing attacks. The tool’s newest iteration, GPT-4, has even more features that could potentially be used for malicious purposes. While the tool has content filters in place, cybercriminals have found ways to work around them, and pressure may need to be put on the tool’s owners to enhance protective measures.
However, AI can also be used for defensive capabilities in the cybersecurity industry. Cybersecurity companies are leveraging AI to find and catalog malicious threats and develop preventative measures. With its enhanced safeguards and ability to detect malicious behavior, AI can ultimately be a powerful asset for organizations. It is important for individuals and organizations to become more vigilant and take proactive measures to prevent misuse of AI tools.
Ultimately, the security threat posed by AI tools like ChatGPT comes from the bad actors who use them for malicious purposes. The responsibility falls on both individuals and organizations to ensure that AI tools are used for positive growth and change, while also mitigating potential risks. By working together and implementing appropriate safeguards, detection methods, and ethical guidelines, we can maximize the benefits of AI while minimizing its potential risks.
Stay up to date with what is happening! Get a first look at news, noteworthy research and worst attacks on AI delivered right in your inbox.
In this research digest, we explore 4 remarkable research papers that delve into diverse aspects of adversarial attacks, from query-free techniques to real-world examples, unveiling the intricate vulnerabilities of advanced ...
