ChatGPT-4o: Exciting features and lingering cybersecurity concerns

The world of Artificial Intelligence is evolving at a breakneck pace, and leading the charge is OpenAI’s ChatGPT. This cutting-edge large language model (LLM) has set new standards for AI, and now, just 18 months after its debut, OpenAI has launched a major update: GPT-4o. While this update brings impressive advancements, it also raises significant cybersecurity concerns that experts are keen to address.

OpenAI introduced GPT-4o, with the “o” standing for “omni,” during a live stream earlier this month in May. This latest version of ChatGPT pushes the boundaries of what AI can do, making it a game-changer in the field. Here’s a closer look at the standout features of GPT-4o:

Enhanced Speed and Multimodality: GPT-4o is faster and more versatile than its predecessors. It excels in processing diverse information formats—text, audio, and visuals—enabling more comprehensive and natural interactions.

Free Tier Expansion: OpenAI is democratising access to AI by offering some GPT-4o features to free-tier users. These include web-based information access during conversations, image discussions, file uploads, and basic enterprise-grade data analysis tools. Paid users still get a broader range of functionalities.

Improved User Experience: GPT-4o can now generate convincingly realistic laughter, potentially bridging the uncanny valley and enhancing user engagement. It’s also adept at interpreting visual input, such as recognising sports on television and explaining the rules, making it a valuable tool for many.

Despite the exciting new features, the launch of GPT-4o has reignited discussions about the potential misuse of AI. Cybersecurity professionals are particularly concerned about how advanced AI models can be exploited by malicious actors.

Since ChatGPT's inception in November 2022, several major organisations—including Amazon, JPMorgan Chase & Co., Bank of America, and others—have restricted or blocked access to the program due to security worries. These concerns are not without merit. In April 2023, Italy became the first country to ban ChatGPT, accusing OpenAI of mishandling user data.

In response to these concerns, OpenAI has implemented several new safety measures in GPT-4o. According to OpenAI, GPT-4o includes "new safety systems to provide guardrails on voice outputs" and extensive post-training data filtering to prevent inappropriate or unsafe responses.

OpenAI follows its internal Preparedness Framework and voluntary commitments to ensure the model's safety. Before the release of GPT-4o, more than 70 external security researchers rigorously tested the model to identify and mitigate potential risks.

In a detailed post on its official website, OpenAI stated that GPT-4o does not score above a “medium risk” in their cybersecurity evaluations. These evaluations cover a range of factors, including cybersecurity, chemical, biological, radiological, and nuclear (CBRN) risks, persuasion, and model autonomy.

OpenAI has gone a step further by engaging over 70 external experts in areas such as social psychology, bias and fairness, and misinformation. These experts conducted extensive red teaming to identify and address risks posed by the new modalities in GPT-4o. Insights from this process have informed the development of robust safety interventions.

Balancing Innovation with Security

As OpenAI continues to innovate with updates like GPT-4o, the balance between advancing AI capabilities and ensuring cybersecurity remains critical. The enhanced features of GPT-4o are undoubtedly exciting and hold great promise for various applications. However, the potential for misuse means that ongoing vigilance and robust safety measures are essential.

OpenAI's commitment to transparency and collaboration with external experts is a positive step toward mitigating risks. As users and developers explore the possibilities of GPT-4o, staying informed and prioritising security will be key to harnessing the full potential of this ground-breaking technology responsibly.

Ultimately, GPT-4o represents a significant leap forward for ChatGPT, offering enhanced capabilities and broader accessibility. Yet, as with any powerful tool, the importance of robust cybersecurity measures cannot be overstated. OpenAI’s proactive approach to safety and security provides some reassurance, but the conversation about AI’s potential risks and benefits is far from over.