Winners & losers: Generative AI has become a double-edged sword for security teams. The same technology that helps uncover and fix vulnerabilities faster than ever also makes it trivially easy to flood inboxes with dubious bug reports. That tension recently pushed Apple to change its bug bounty program in a way that ended up delaying disclosure of a genuinely serious exploit.
Security researchers at Bynario recently told the Financial Times that Apple's new bug bounty policy held up its efforts to report dozens of vulnerabilities. Among them was a privilege escalation exploit worth up to $200,000 on the black market.
Apple confirmed to the FT that it has since contacted Bynario, and that it capped how many reports a researcher can have open at once in response to a flood of AI-generated claims. Once a researcher hits that cap, they face a 30-day cool-off period before they can file again, though they can request a higher quota through the company's security portal at any time.
The Cupertino giant is grappling with a problem that's been building for a couple of years. Curl's security team, for instance, has been sounding the alarm on AI slop since early 2024; by 2025, confirmed-vulnerability rates on its bug bounty program had fallen below 5%, down from more than 15% before the AI-slop wave hit.
Enforcing bug-report quotas can backfire, though, because AI has also increased the number and quality of legitimate bug reports. Last year, Bynario's seven-person team reported eight exploits to Apple; this year, it used ChatGPT to uncover more than 50 in just three weeks. The researchers tried to report five of them to Apple, but the company's new quota system initially blocked the disclosures.
One of those exploits, tracked as CVE-2026-43760, targeted a legacy code path in macOS Screen Sharing's VNC password authentication, a holdover for older VNC clients that don't use full macOS credentials. It let an authenticated VNC viewer read protected files outside its authorized scope, then escalate that into a working exploit capable of executing commands with root privileges.
Notably, Bynario's researchers say the technique worked without triggering memory corruption at all, putting it outside the reach of Apple's Memory Integrity Enforcement system, which is designed specifically to catch memory-corruption attacks.
The core problem is that even though Apple and other companies use AI to help sift through bug reports, verifying a submission still takes far more human time and effort than generating one does. Still, recent patches show the upside of AI-assisted bug hunting.
Apple's latest security update, released in late July, addressed nearly 200 issues across iPhones, Safari, the App Store, the macOS kernel, and numerous other Apple products. Chrome's last two versions fixed more than 1,000 vulnerabilities combined – that's more than the previous 23 releases put together(!). That acceleration has already prompted Google to speed up its Chrome release cadence.