Author: @riskvisions

Can We Trust AI with Nuclear Weapons?

The integration of Artificial Intelligence (AI) into nuclear command and control systems presents a complex challenge, demanding a nuanced approach to risk management that goes beyond the simplistic “human-in-the-loop” model. https://europeanleadershipnetwork.org/report/ai-and-nuclear-command-control-and-communications-p5-perspectives/

While maintaining human oversight is crucial, it is not a foolproof safeguard against unintended escalation. The limitations of current AI models, such as hallucinations, opacity, and susceptibility to cyberattacks, can lead to flawed predictions, skewed decision-making, and compromised system integrity. Furthermore, the rapid evolution of AI means that new, unforeseen risks may emerge, making a static “human-in-the-loop” approach insufficient. https://warontherocks.com/2024/12/beyond-human-in-the-loop-managing-ai-risks-in-nuclear-command-and-control/https://warontherocks.com/2024/12/beyond-human-in-the-loop-managing-ai-risks-in-nuclear-command-and-control/

Instead, a more robust framework is needed, one that focuses on the overall safety performance of the AI-integrated system. This framework should establish a quantitative threshold for the maximum acceptable probability of an accidental nuclear launch, drawing lessons from civil nuclear safety regulations.

A performance-based approach, similar to that used in civil nuclear safety, would define specific safety outcomes without prescribing the exact technological means to achieve them. This would allow for flexibility in adapting to evolving AI capabilities while ensuring that the risk of accidental launch remains below an acceptable level.

The adoption of probabilistic risk assessment techniques, which quantify the likelihood of various accident scenarios, would provide a more comprehensive understanding of the risks involved in AI integration. This quantitative approach would enable policymakers to make informed decisions about the acceptable levels of AI integration in nuclear command and control systems.

The international community must move beyond mere declarations of human control and engage in a deeper discussion about AI safety in the nuclear domain. This discussion should focus on establishing quantifiable safety objectives and developing a performance-based governance framework that can adapt to the evolving risks of AI.

2 March 2025
DeepSeek: A New Challenger in the AI Race

The global AI landscape is undergoing a significant shift with the emergence of DeepSeek, a Chinese company that has released powerful open-source AI models. DeepSeek’s models, DeepSeek-V3 and DeepSeek-R1, rival the performance of leading American models like OpenAI’s ChatGPT and Anthropic’s Claude, but at a fraction of the cost. This has allowed developers and users worldwide to access cutting-edge AI technology, posing a potential threat to U.S. leadership in the field. https://www.foreignaffairs.com/china/real-threat-chinese-ai

The Open-Source Advantage

DeepSeek’s models are open-source, meaning anyone can download, modify, and build upon them. This stands in contrast to the predominantly proprietary models offered by American AI companies. Open-source software has historically fostered innovation and collaboration, allowing for rapid development and enhanced security. DeepSeek’s open-source approach has enabled it to achieve remarkable performance despite facing export controls that limit its access to advanced chips.

The Chinese Influence

While DeepSeek’s open-source models offer significant advantages, they also raise concerns about potential Chinese government influence. Beijing has implemented regulations requiring Chinese-made LLMs to align with the “core values of socialism” and avoid disseminating sensitive information. This censorship is evident in DeepSeek’s models, which avoid or provide skewed answers on topics deemed sensitive by the Chinese government.

The Chip Advantage

The rise of DeepSeek also has implications for the AI chip market. Currently, the U.S. dominates the AI chip market, with Nvidia’s GPUs powering the majority of AI workloads. However, DeepSeek’s ability to run its models on less advanced hardware, such as Huawei’s Ascend chips, could shift the market towards Chinese-made chips. This could potentially erode the West’s chip advantage and give China greater control over the AI supply chain.

A Call for Action

The emergence of DeepSeek highlights the need for the United States to reassess its AI strategy. While continuing to invest in frontier AI systems, the U.S. must also prioritize the development and support of open-source AI models. This includes increasing funding for open-source AI initiatives, creating incentives for companies to release open-source models, and fostering a robust open-source AI ecosystem.

The AI race is not just about technological advancement; it’s also about shaping the future of the global AI landscape. The United States must act decisively to ensure that it remains a leader in the AI race and that the development and deployment of AI technologies align with democratic values.

1 March 2025
AI Disinformation: A Growing Threat to Truth and Trust

The rise of artificial intelligence (AI) has brought about many benefits, but it has also created new challenges. One of the most concerning of these challenges is the use of AI to generate and spread disinformation. AI-powered tools can now create highly realistic fake videos, audio recordings, and images, making it easier than ever to spread false information and manipulate public opinion. In a recent summit in France, (https://www.elysee.fr/en/sommet-pour-l-action-sur-l-ia/trust-in-ai) french President Emmanuel Macron cited a “need for rules” to govern artificial intelligence, in an apparent rebuff to US Vice President JD Vance, who had criticized excessive regulation.

The use of AI-generated disinformation is already having a significant impact on society. In recent elections, deepfakes have been used to discredit candidates and influence voters. AI-generated fake news articles have been used to spread false narratives and sow discord. And AI-powered bots have been used to amplify disinformation and manipulate online conversations.

The threat of AI disinformation is only going to grow in the years to come. As AI technology becomes more sophisticated, it will become even easier to create realistic fake content. This could have a devastating impact on our ability to trust the information we see online.

There are a number of things that can be done to combat the threat of AI disinformation. One is to raise awareness of the issue. The more people know about AI disinformation, the less likely they are to be fooled by it. Another is to develop tools that can help people identify fake content. A third is to work with social media companies to ensure that they are taking steps to prevent the spread of disinformation on their platforms.

The fight against AI disinformation is a critical one. If we fail to address this challenge, we risk undermining our ability to trust the information we see online and make informed decisions about our lives. Sources and related content

26 February 2025
AI-Enabled Cyber Warfare: A New Dimension of Conflict

The rapid advancement of AI is not just transforming industries and economies; it’s also changing the landscape of warfare. One of the most concerning trends is the use of AI to develop and deploy cyber weapons, ushering in a new dimension of conflict with potentially devastating consequences.

AI-powered cyberattacks can be more sophisticated, targeted, and destructive than traditional cyberattacks. AI algorithms can be used to automate the process of identifying and exploiting vulnerabilities in software and networks, making it possible to launch attacks with unprecedented speed and scale. AI can also be used to evade detection by learning to mimic normal network traffic or by adapting to changing defenses.

But there’s another layer to this emerging threat: autonomous AI agents. These agents can interact with the environment, make decisions, and take action without human intervention. In the context of cyber warfare, this means AI agents could potentially identify and exploit vulnerabilities, launch attacks, and even adapt to defenses – all without direct human control. https://www.fastcompany.com/91281577/autonomous-ai-agents-are-both-exciting-and-scary

The consequences of AI-enabled cyberattacks could be severe. Attacks on critical infrastructure, such as power grids, financial systems, and healthcare networks, could disrupt essential services and cause widespread chaos. AI-powered cyber espionage could lead to the theft of sensitive data, such as intellectual property and government secrets. And AI-enabled disinformation campaigns could be used to manipulate public opinion and sow discord.

Defending against AI-powered cyberattacks, especially those launched by autonomous agents, is a major challenge. Traditional cybersecurity tools and techniques may not be effective against AI-enabled attacks. AI systems can learn to evade detection and adapt to changing defenses, making it difficult to develop effective countermeasures. The development of new AI-powered cybersecurity tools and the training of cybersecurity professionals in AI-related skills are crucial to address this challenge.

International cooperation is essential to mitigate the risks of AI-enabled cyber warfare. The development and deployment of AI cyber weapons is a global threat that requires a coordinated response. Countries need to work together to develop norms and standards for the responsible use of AI in cyber operations, as well as to share information about AI threats and vulnerabilities.

The threat of AI-enabled cyber warfare is real and growing. It is crucial that we take steps now to mitigate the risks and ensure that AI is used for good, not for harm.

23 February 2025
The Responsibility Gap: Why AI Governance Needs to Catch Up

The rapid pace of AI innovation is exciting, but it also presents new challenges for safety and responsibility. As AI systems become more sophisticated, they can be used for harmful purposes, or they can fail in unexpected ways. This is why AI governance needs to catch up with the pace of innovation.

According to a recent survey of C-suite executives, there is a “responsibility gap” between the rapid pace of AI innovation and the slower development of effective governance https://us.nttdata.com/en/news/press-release/2025/february/ntt-data-report-exposes-the-ai-responsibility-crisis . Business leaders are calling for more clarity on regulation, and many are worried about the security risks of AI.

One of the biggest challenges is that there is no one-size-fits-all approach to AI governance. The risks and benefits of AI vary depending on the specific application. This means that policymakers need to take a nuanced approach, considering the specific risks and benefits of each AI system.

Another challenge is that AI is a global technology. This means that international cooperation is needed to develop effective governance frameworks. However, international cooperation can be difficult to achieve, especially in the face of competing national interests.

Despite these challenges, there are a number of things that can be done to improve AI governance. One is to invest in research on AI risks and risk management. This research can help policymakers to develop more effective governance frameworks. Another is to develop international standards for AI safety and responsibility. These standards can help to ensure that AI is developed and used in a way that benefits humanity.

AI is a powerful technology with the potential to transform our world in many ways. However, it is important to ensure that AI is developed and used responsibly. By taking a proactive approach to AI governance, we can help to ensure that AI benefits humanity as a whole.

What do you think? How can we close the responsibility gap and ensure that AI is developed and used responsibly?

17 February 2025
The Osama bin Laden Scenario: Could AI Be the Next Weapon of Mass Destruction?
In the wake of the recent AI summit in Paris, former Google CEO Eric Schmidt expressed a chilling concern: could AI become the next weapon of mass destruction? https://www.politico.eu/article/ex-google-boss-eric-schmidt-fears-ai-risks-could-lead-to-a-bin-laden-scenario/

Schmidt cautioned that AI falling into the wrong hands could pose extreme risks, evoking a potential “Osama bin Laden scenario.” In this scenario, terrorists or rogue states could use AI to cause harm to innocent people.

This is not mere science fiction. The International Scientific Report on the Safety of Advanced AI, published in January 2025, details the many ways AI could be misused. The report highlights AI’s potential for malicious use, including cyberattacks, the development of chemical and biological weapons, and the spread of disinformation.

The concern over AI misuse is not limited to a few experts. In a recent survey, 90% of respondents expressed concern about the potential for AI to be used for harmful purposes. This public anxiety is not unfounded. AI is a dual-use technology, meaning it can be used for both beneficial and harmful purposes.

The potential for AI to be used as a weapon of mass destruction is a grave concern. AI could be used to design and deploy cyberattacks that could cripple critical infrastructure, such as power grids and financial systems. AI could also be used to develop and deploy chemical and biological weapons that could cause mass casualties.

The threat of AI terrorism is real and growing. As AI becomes more sophisticated and accessible, the risk of it being used for malicious purposes will only increase. It is crucial that we take steps now to mitigate these risks and ensure that AI is used for good, not for harm.

What Can Be Done?

There is no single solution to the threat of AI terrorism. However, there are several steps that can be taken to mitigate the risks.
- Increase government oversight: Governments need to increase their oversight of AI development and use. This includes regulating the development and deployment of AI systems, as well as monitoring the use of AI by both state and non-state actors.
- Invest in AI safety research: More research is needed to understand the risks of AI and how to mitigate them. This includes research on AI safety, AI ethics, and AI governance.
- Promote international cooperation: International cooperation is essential to address the global threat of AI terrorism. This includes sharing information about AI risks and coordinating efforts to mitigate them.
- Educate the public: The public needs to be educated about the risks of AI and how to protect themselves from AI-enabled attacks. This includes educating people about AI safety, AI ethics, and AI governance.
The threat of AI terrorism is real and growing. However, by taking steps now to mitigate the risks, we can ensure that AI is used for good, not for harm.
16 February 2025
The International AI Safety Report: A Wake-Up Call for Risk Professionals
The latest International AI Safety Report is a must-read for anyone working in risk management. https://www.gov.uk/government/publications/international-ai-safety-report-

This landmark report, compiled by 96 AI experts worldwide, offers a stark assessment of the evolving AI landscape and its potential impact on society.

Here are the key takeaways that risk professionals need to know:

AI is evolving at an unprecedented pace. Just a few years ago, AI models struggled to generate coherent text. Today, they can write complex computer programs, create photorealistic images, and even engage in nuanced conversations. This rapid progress shows no signs of slowing down, with experts predicting further significant advancements in AI capabilities in the coming years.

This rapid evolution brings new and complex risks. The report highlights several existing and emerging risks associated with AI, including:
- Malicious use: AI can be used to create harmful deepfakes, manipulate public opinion, launch sophisticated cyberattacks, and even facilitate the development of bioweapons.
- Malfunctions: AI systems can be unreliable, biased, and prone to errors, potentially leading to harmful consequences in critical domains like healthcare and finance.
- Systemic Risks: The widespread adoption of AI could lead to significant labor market disruptions, exacerbate global inequalities, and erode privacy.
Risk management is struggling to keep up. The report emphasizes that current risk management techniques are often insufficient to address the complex and evolving risks posed by AI. This “evidence dilemma” requires policymakers and risk professionals to make difficult decisions with limited information and often under pressure.

The need for a proactive and comprehensive approach. The report calls for a more proactive and comprehensive approach to AI risk management, involving:
- Increased investment in AI safety research.
- Greater collaboration between AI developers, policymakers, and civil society.
- The development of robust risk assessment frameworks and mitigation strategies.
The International AI Safety Report serves as a wake-up call for risk professionals. It’s a reminder that the AI revolution is not just about technological advancement, but also about managing the risks that come with it. By understanding the key takeaways of this report and embracing a proactive and comprehensive approach to risk management, we can help ensure that AI benefits society while minimizing its potential harms.
2 February 2025
Unlocking the Power of AI in Risk Management: 5 Key Advantages

The world is changing rapidly, and with it, the risks businesses face are becoming more complex and unpredictable. Traditional risk management approaches are often struggling to keep up. Enter Artificial Intelligence (AI), a game-changer with the potential to revolutionize how we identify, assess, and mitigate risks.

I believe in harnessing the power of AI to enhance risk management strategies. Here are five key advantages of incorporating AI into your risk management framework:

1. Enhanced Efficiency and Productivity:

Tired of manual, time-consuming risk assessments? AI and Machine Learning (ML) can automate data analysis, freeing up your team to focus on strategic decision-making. Imagine processing vast amounts of data with minimal human intervention, leading to faster and more efficient risk identification and assessment. This applies to both back-office functions like regulatory compliance and front-office operations like customer risk profiling.

2. Superior Forecasting Accuracy:

Traditional risk forecasting methods often rely on historical data and assumptions that may not hold true in a dynamic environment. AI-powered techniques can analyze complex patterns and provide more accurate predictions, enabling you to anticipate and mitigate potential threats proactively. This is particularly valuable for credit risk modeling, where AI can explore data and predict crucial credit risk characteristics.

3. Cost Reduction:

Implementing AI in risk management can lead to significant cost savings. By automating processes and streamlining operations, AI/ML solutions can reduce operational, regulatory, and compliance costs for financial institutions and businesses across various sectors.

4. Deeper Customer Insights:

AI/ML solutions can generate vast amounts of accurate data in a timely manner. This allows you to develop a comprehensive understanding of your customers, including their risk profiles and behaviors. These insights can be used to implement targeted risk mitigation strategies and reduce potential losses.

5. Improved Decision Making:

AI’s ability to analyze data, predict outcomes, and generate actionable insights empowers you to make better decisions across all levels – strategic planning, operations, and daily tactical choices. This is particularly valuable for investment and business-related decisions, enabling you to navigate uncertainty with confidence.

The Future of Risk Management is AI-Driven:

AI is no longer a futuristic concept; it’s a present-day reality that’s transforming risk management. By embracing AI-powered solutions, your organization can gain a competitive edge, reduce costs, and make better decisions in the face of uncertainty.

Disclaimer: While AI offers significant advantages for risk management, it’s important to be aware of potential risks associated with its implementation. Future blog posts will explore these risks and discuss strategies for mitigating them

2 February 2025
AI Fighting Fire with Fire: How AI Techniques are Revolutionizing Risk Management
Artificial intelligence (AI) isn’t just changing the world around us; it’s also transforming how we manage the risks that come with those changes. With the rapid advancement of technology, AI now offers a powerful arsenal of techniques that can be directly applied to identify, analyse, and mitigate risks across various sectors. From finance to healthcare, the potential applications of AI in risk management are vast and varied.

I’m excited about the potential of these techniques to enhance risk management strategies significantly. Let’s explore some of the key AI methods that are reshaping the field, along with real-world applications that illustrate their impact:

1. Machine Learning (ML): The Pattern Recognition Powerhouse

Machine learning algorithms excel at recognising patterns in vast amounts of data. In risk management, this translates to several practical applications:
- Predictive Modeling: ML can analyze historical data to identify trends and predict future events, such as credit defaults, fraud, or operational disruptions.
- Anomaly Detection: By learning normal behavior patterns, ML can quickly identify unusual activities that may indicate potential risks, such as cyberattacks or insider threats.
- Classification: ML can classify risks based on their severity, likelihood, and potential impact, aiding in prioritization and response planning.
2. Deep Learning: Delving Deeper into Complexity

Deep learning, a subset of ML, uses artificial neural networks to analyse complex, unstructured data like images, text, and audio. This opens up new possibilities for risk management, allowing organisations to tap into previously untapped data sources:
- Unstructured Data Analysis: Deep learning can extract valuable insights from sources like social media posts, news articles, and customer reviews to identify emerging risks or sentiment shifts.
- Image and Video Analysis: In industries like security and surveillance, deep learning can analyze images and videos to detect anomalies or potential threats in real-time.
- Natural Language Processing (NLP): NLP can be used to analyze text-based data like contracts, legal documents, and customer communications to identify potential risks or compliance issues.
3. Natural Language Processing (NLP): Understanding the Human Factor

NLP focuses on enabling computers to understand and process human language. In risk management, this can be applied in various ways that benefit organisations:
- Sentiment Analysis: Analyze customer feedback, social media posts, and news articles to gauge public sentiment and identify potential reputational risks.
- Fraud Detection: Detect fraudulent activities by analyzing communication patterns and identifying suspicious language or behavior.
- Compliance Monitoring: Monitor internal and external communications for compliance with regulations and ethical guidelines.
4. Reinforcement Learning: Learning by Doing

Reinforcement learning involves training AI agents to make decisions in complex environments through trial and error. This technique can be applied to various risk management scenarios:
- Dynamic Risk Mitigation: Develop AI systems that can adapt and adjust risk mitigation strategies in response to changing conditions.
- Optimized Decision-Making: Train AI agents to make optimal decisions in complex situations, such as portfolio management or resource allocation.
- Automated Response: Develop AI systems that can automatically respond to certain types of risks, such as cybersecurity threats or operational incidents.
The Future of AI in Risk Management

These AI techniques are just the tip of the iceberg. As AI technology continues to evolve, we can expect even more innovative applications in risk management. By embracing these advancements, organisations can gain a significant advantage in proactively mitigating risks and building resilience in an increasingly uncertain world. Moreover, the integration of AI in risk management is becoming essential for businesses that wish to thrive in the face of challenges and uncertainties.

Stay tuned for future blog posts that will explore specific use cases and applications of AI in risk management across various industries. We’ll discuss how different sectors can leverage AI techniques to enhance their risk management frameworks and ensure long-term sustainability.
31 January 2025
Five Urgent AI Risks to Watch Out For
Artificial intelligence (AI) is rapidly changing the world around us, but this rapid progress comes with significant risks, particularly AI risks. Organizations, individuals, and society as a whole need to be aware of these AI risks and take steps to mitigate them.

Here are five key AI risks, based on information from the “International AI Safety Report”:

Understanding AI risks is essential for everyone in today’s technology-driven landscape.

Understanding AI Risks
1. Bias and unreliability in data and models: AI systems can produce inaccurate or biased results due to the data they are trained on. This can lead to discriminatory outcomes and erode trust in AI. For example, an AI system used for loan applications might unfairly discriminate against certain groups if the training data reflects historical biases.
2. Lack of transparency and explainability: Many AI systems are “black boxes,” making it difficult to understand how they make decisions. This lack of transparency can hinder accountability and make it harder to identify and correct errors or biases. For instance, if a self-driving car makes an accident, it might be difficult to determine why without a clear understanding of the AI’s decision-making process.
3. Security risks and vulnerability to attacks: AI models can be vulnerable to various security threats, including theft, manipulation, and adversarial attacks. Attackers could exploit these vulnerabilities to steal sensitive data, disrupt critical systems, or spread misinformation. For example, hackers could manipulate an AI-powered medical diagnosis system to provide false diagnoses, putting patients at risk.
4. Operational risks: AI systems can experience performance degradation over time due to changes in data or relationships between data points. Integration challenges with existing IT infrastructure, sustainability concerns, and a lack of oversight can also create problems. For example, an AI system used for fraud detection might become less effective as new fraud patterns emerge, potentially leading to financial losses.
5. Existential and societal risks: Some experts believe that advanced AI could pose existential risks to humanity, while others are concerned about its potential to exacerbate existing societal problems. AI-driven automation could lead to job displacement, and AI systems could be used to manipulate individuals or spread misinformation. There are also concerns that AI could be used to develop autonomous weapons systems or to create other dangerous technologies.
These AI risks highlight the urgent need for robust risk management frameworks and mitigation strategies. Organizations and policymakers need to proactively address these AI risks to ensure that AI is developed and used responsibly.

To learn more about these risks and potential mitigation strategies, you can check out the following resources:
- International AI Safety Report: https://www.gov.uk/government/publications/international-ai-safety-report-2025
- OECD AI Policy Observatory: https://www.oecd.org/en/topics/policy-issues/artificial-intelligence.html
By staying informed and taking a proactive approach to risk management, we can harness the benefits of AI while mitigating its potential harms. Sources and related content
31 January 2025