Artificial intelligence is finding software vulnerabilities so quickly that Apple can no longer review them all at once. The company has introduced a cap on how many security reports individual researchers can have open simultaneously after a surge in AI-assisted submissions overwhelmed its review pipeline. It is one of the clearest signs yet that AI is not just changing how software gets built — it is changing how it gets broken, and how quickly.
The challenge is not simply volume. AI-powered security tools can generate reports at a pace no human team can match, but many submissions contain theoretical vulnerabilities rather than real ones — problems that cannot be exploited in practice, or that rest on incorrect assumptions made by the AI during analysis. Apple still relies on human experts to verify every submission before deciding whether a vulnerability requires a patch. Even with AI now helping to prioritise the queue, the final determination remains manual — and the backlog keeps growing.
What AI-Assisted Security Research Looks Like in Practice
Security company Bynario offered a concrete illustration of the new pace. Using its Atlas platform, powered by GPT-5.5, the company identified more than 50 potential macOS vulnerabilities within three weeks. Among them was a privilege escalation chain capable of giving an attacker complete control over a Mac.
One vulnerability, affecting macOS Screen Sharing, allowed an authenticated VNC viewer to access protected information and create files with root-level privileges under specific conditions. The flaw required Screen Sharing or Remote Management to be enabled alongside legacy VNC password authentication — a configuration that exists in real-world environments. Apple assigned the issue the identifier CVE-2026-43760 and released a fix in macOS Tahoe 26.6.
Critically, Bynario submitted a working proof of concept rather than a theoretical description, allowing Apple’s engineers to reproduce the issue and develop the patch more efficiently. That distinction matters: demonstrated exploits with technical evidence can be processed far more quickly than hypothetical attack paths, and Apple has updated its bug bounty programme to reflect this, placing greater emphasis on evidence that an exploit can successfully reach protected parts of its operating systems.
Apple’s maximum bug bounty reward now exceeds $5 million for the most serious exploit chains. The programme also includes Target Flags — a mechanism that helps researchers demonstrate a vulnerability can access sensitive system areas rather than simply describing a theoretical weakness.
AI Is Already Producing Real Patches
Apple’s own security advisories confirm that AI is already contributing to genuine security improvements. The company has credited researchers working with Anthropic’s Claude for discovering a kernel vulnerability, while OpenAI Codex Security has assisted in identifying multiple issues affecting the WebKit browser engine that underpins Safari. These are not theoretical contributions — they resulted in official patches.
The pattern emerging is one where AI handles the broad, fast scan across large codebases while human expertise validates which findings represent real threats. Rather than replacing security researchers, AI tools are amplifying what they can examine and how quickly, generating more leads than ever before for human experts to follow up.
The problem, as Apple is discovering, is that more leads require more reviewers — and human review does not scale at the same rate as AI discovery.
What This Means for Mac Users
For everyday Mac users, the most practical takeaway is straightforward: install software updates as soon as they are available. Apple’s ability to process and patch vulnerabilities depends on a functioning review pipeline, and the timeliness of that process is under more pressure than at any previous point. Updates represent the primary defence against vulnerabilities that have already been discovered and reported — whether by human researchers or AI.
Apple must also navigate a difficult balance in how it manages submissions. Restricting them too aggressively risks discouraging legitimate security researchers from reporting important discoveries. Accepting unlimited AI-generated reports risks burying genuine threats under a pile of false positives. Neither outcome serves users.
The broader implication extends well beyond Apple. AI-powered vulnerability research is becoming standard across the software security industry, and every major technology company running complex operating systems faces a version of the same challenge. The pace of discovery is accelerating faster than the pace of review — and building the processes to close that gap is now one of the more urgent problems in software security.
Stay informed. Subscribe to the JournalTodays Newsletter for the latest Apple news, cybersecurity coverage, and technology updates delivered straight to your inbox.
