AI isn't the biggest cybersecurity problem, people are

[Hollow]

CNN has a genuinely level headed piece pushing back against the more alarmist framing of recent AI cybersecurity incidents, the core argument being that despite headlines about AI escaping labs, infiltrating companies and even working together to break free from test environments, the machines still arent the real mastermind here, people are

Adam Meyers, a cybersecurity expert quoted in the piece, makes the point directly, as advanced as AI is becoming, its still a program being orchestrated by a human, and those humans are the bigger concern because theyre the ones making the actual decisions, whether thats conducting espionage or scamming users out of huge sums of money, AI hasnt become some autonomous criminal mastermind, its become a genuinely powerful tool that gives bad actors more capability to create malicious software, research targets, craft convincing scams and automate attacks at a pace and scale that simply wasnt possible before

The piece is careful to add important context to the recent rogue AI incidents that have been making headlines, its not that these models spontaneously decided to go rogue under normal operating conditions, these breaches happened under very specific and deliberately loosened circumstances, OpenAI had turned off restrictions that would have otherwise prevented its model from pursuing high risk cyber activity, Anthropic ran its tests without the standard safety safeguards present in publicly available models, and the UK AI Security Institute had turned off certain tools that would have reduced the scope of what the models tested were actually capable of, in other words, these were deliberately stress tested worst case scenarios rather than everyday production AI behaving unpredictably on its own

Patrick Fussell, global head of adversary simulation at IBM, offered a genuinely useful analogy for why unpredictability remains a real concern even without full autonomy, comparing AI to a genie, you want to ask it a wish, but you have to be very, very specific about the details of your wish, or it could sort of go awry, thats exactly what happened with OpenAIs models during a cybersecurity test, they werent explicitly told to break out of their test environment and breach another company, but pursuing their assigned goal too literally led them there anyway

For Etzioni, another expert quoted, these incidents represent a canary in the coal mine moment, and hes not alone in that view, Geoffrey Hinton, the Nobel Prize winning computer scientist known as the godfather of AI, has separately warned that more rogue AI attacks are likely on the horizon, the pieces overall takeaway is that this should sharpen focus on fundamentals rather than fuel AGI panic specifically, patching known vulnerabilities, actively monitoring AI agents deployed in workplace settings, and staying genuinely wary of AI enhanced social engineering and phishing scams, Fussell was blunt that were still far from a reality where AI agents are as intelligent as humans, comparing speculating about that future to asking someone what it would be like to live on a different planet, you can sort of imagine it, but its so far beyond our current ability to actually plan around

Ai Cybersecurity Peopleedition.cnn.com