The cybersecurity industry is fascinated by artificial intelligence (AI), which has the ability to completely change how security and IT teams respond to cyber crises, breaches, and ransomware assaults.
However, it's important to have a realistic knowledge of the capabilities and limitations of AI, and there are a number of obstacles that prevent the technology from having an instant, profound influence on cybersecurity. In this article, we examine the limitations of AI in dealing with cybersecurity issues while emphasising the part played by organisations in empowering resilience and data-driven security practices.
Inaccuracy
The accuracy of its output is one of AI's main drawbacks in cybersecurity. Even though AI systems, like generative pre-trained transformers like ChatGPT, can generate content that is in line with the internet's zeitgeist, their answers are not necessarily precise or trustworthy.
AI systems are excellent at coming up with answers that sound logical, but they struggle to offer accurate and trustworthy solutions. Given that not everything discovered online is factual, relying on unfiltered AI output can be risky.
Recovery actions' complexity
Recovery following a cyber attack often involves a complex series of procedures across multiple systems. IT professionals must perform a variety of actions in order to restore security and limit the harm. Entrusting the entire recovery process to an AI system would necessitate a high level of confidence in its dependability.
However, existing AI technology is too fragile to manage the plethora of operations required for efficient cyberattack recovery. Directly linking general-purpose AI systems to vital cybersecurity processes is a huge problem that requires extensive research and testing.
General intelligence vs. General knowledge
Another distinction to make is between general knowledge and general intelligence. While AI systems like ChatGPT excel at delivering general knowledge and generating text, they lack general intelligence. These systems can extrapolate solutions based on prior knowledge, but they lack the problem-solving abilities involving true general intelligence.
While dealing with AI systems via text may appear to humans to be effective, it is not consistent with how we have traditionally interacted with technology. As a result, current generative AI systems are limited in their ability to solve complex IT and security challenges.
Making ChatGPT act erratically
There is another type of threat that we must be aware of: the nefarious exploitation of existing platforms. The possibility of AI being "jailbroken," which is rarely discussed in the media's coverage of the field, is quite real.
This entails giving text commands to software like ChatGPT or Google's Bard in order to circumvent its ethical protections and set them free. By doing this, AI chatbots get transformed into powerful assistants for illegal activities.
While it is critical to avoid the weaponization of general-purpose AI tools, it has proven extremely difficult to regulate. A recent study from Carnegie Mellon University presented a universal jailbreak for all AI models, which might create an almost limitless amount of prompts to circumvent AI safeguards.
Furthermore, AI developers and users are always attempting to "hack" AI systems and succeeding. Indeed, no known universal solution for jailbreaking exists as of yet, and governments and corporations should be concerned as AI's mass adoption grows.
