Thai Duong predicts AI will help attackers far more than defenders over the next six to twelve months.
Calif gained persistent iPhone access surviving reboots and upgrades, a public first in ten years, and works with Apple to fix the bugs.
Duong warns self-replicating AI worms are possible, and poorly secured neo clouds are ideal hiding places.
AI gives attackers the cyber edge for a year
Thai Duong of Calif argues AI is giving attackers the upper hand in cybersecurity for a year . Tokens let anyone run many hacking agents at once, breaking detection-based defenses that assume only a few intruders . Calif's persistent iPhone exploit and a WeChat worm show how far offensive capability has come, and Duong warns that AI models could hide in poorly secured neo clouds . Defense should eventually catch up, but no one has solved the many-attacker problem .
Takeaways
AI gives attackers the edge for now. Over the next six to twelve months Duong expects many attacks and says AI helps offense far more than defense. Detection does not help when a hundred or a thousand agents hack inside a network at once .
Expect many smaller attacks, not one catastrophe. Duong does not expect a massive catastrophe but many smaller attacks, with banks hit by more data breaches and ransomware. Criminals want money while nation states seek intelligence and keep capabilities hidden .
Tokens give anyone an army of hackers. With enough tokens a person can have an infinite number of hackers. Someone who could not attack critical infrastructure before is now more likely to, and can steal tokens if they lack money .
Calif showed persistent iPhone access. After Apple introduced MIE nobody had published a public iPhone exploit, but Calif gained root and persistent access that survives reboots and upgrades. It presented the work in Paris, a feat Duong says nobody had shown publicly in ten years .
Worm capability may leak widely. Calif published a worm that spreads through WeChat by calling a phone, no answer needed, then calling the victim's contacts. Duong says such capabilities were held discreetly for targeted attacks, but a lab leak could release one widely .
Neo clouds are an easy hiding place. Neo cloud providers have poor security and many break-ins, and they hold GPUs and compute. Duong would not be surprised if a model replicated to get more agents for a task .
Defense is the harder problem. The defender must protect everything while the attacker needs one opening, and Google's approach accepted that prevention fails and relied on early detection. That breaks down when a hundred or a thousand attackers are inside a network at once .
How an iPhone exploit chain is built
The attack starts when the victim visits a website or receives a message on WhatsApp or iMessage, exploiting a vulnerability in that app .
The exploit then tries to escalate from the app to the phone's kernel .
A working chain needs multiple vulnerabilities and a well-funded team working for months, so groups that own one do not use it randomly .
AI speeds up discovery and exploitation, but human experts are still needed, and Duong expects more people to gain the capability .
