Apple has quietly placed a limit on the number of open security reports that researchers can submit as AI-assisted vulnerability hunting overwhelms its review queue. The move comes as automated tools generate large volumes of possible macOS and Safari flaws, some of them serious and many of them speculative. Every submission still needs human verification, but the pressure of sorting through a growing stack of AI-assisted findings has forced Apple to rethink how it processes incoming bug reports.

The new bottleneck: triage, not discovery

For years, the hardest part of finding security vulnerabilities was uncovering them in the first place. Researchers spent weeks reading assembler code, tracing kernel paths, and experimenting with memory corruption. Apple's bug bounty program was designed for a slower world in which a single researcher could responsibly report a handful of issues at a time. AI has changed that dynamic. Tools powered by large language models can scan codebases, reason about privilege boundaries, and suggest attack chains quickly enough to generate dozens of possible findings in days.

According to reports, Apple has now limited the number of bug reports that a researcher can keep open at the same time. The cap prevents any single researcher or automated pipeline from flooding Apple's bug bounty system with hundreds of submissions overnight. However, it also creates a new problem: Apple needs enough human analysts to evaluate the findings that do come in. The company has reportedly started using AI to help triage the backlog, but automated triage is not the same as automated verification. Each report still has to be checked against a real system, reproduced in a controlled environment, and assessed for its actual security impact.

The core issue is that AI can produce a convincing description of a vulnerability without that vulnerability being real. Some submissions describe hallucinated APIs, impossible call paths, or purely theoretical risks that have no effect in practice. Others uncover vulnerabilities serious enough to require patches. The difficulty for Apple is distinguishing between the two quickly enough to respond to genuine threats.

Real vulnerabilities among the noise

One security firm, Bynario, told reporters that it found more than 50 possible macOS flaws in three weeks using its AI-driven platform. The findings included a privilege-escalation chain that could give an attacker full control of a Mac. Privilege escalation is especially dangerous because it allows a process with limited user privileges to gain root-level access. In the wrong hands, a chain like that could be used to install persistent malware, bypass system protections, and read sensitive files from other users.

Bynario has already shown that its system can produce more than automated guesswork. Its Atlas platform used GPT-5.5 to uncover a macOS Screen Sharing flaw that let an authenticated VNC viewer access protected data and create files with root privileges. The attack required Screen Sharing or Remote Management to be enabled, along with legacy VNC password access. Apple assigned the issue CVE-2026-43760 and patched it in macOS Tahoe 26.6. Bynario also demonstrated how the flaw could be extended to run commands as root, giving Apple a working exploit to investigate rather than another vague warning generated from a code scan.

That distinction matters. A proof-of-concept exploit gives Apple's security teams a concrete starting point for understanding the bug. A vague AI-generated report, by contrast, may mention the right functions and the right data flow but fail to account for sandbox restrictions, code signing, or other mitigating controls. Apple has to spend time proving that the reported path is actually reachable before it can decide whether a patch is needed.

Why Apple needs the same AI

Apple's recent security advisories have credited researchers working with AI models for several serious issues. Researchers using Anthropic's Claude were credited for a kernel vulnerability, and OpenAI's Codex Security has helped identify multiple WebKit issues. These are not marginal findings. Kernel and WebKit vulnerabilities often occupy the highest tiers of Apple's security response because they can be exploited in remote attacks or to break out of the browser sandbox.

AI-assisted research is already contributing to fixes shipped for macOS and Safari. That makes the reporting backlog a delicate problem. Restricting submissions too aggressively could delay useful discoveries, while leaving the gates open risks burying Apple's team under convincing-looking nonsense. Apple's recent choices suggest it is trying to strike a balance by limiting the number of open reports but not the number that can be submitted over time. Researchers who close old submissions or see them resolved can continue to open new ones, which keeps the pipeline moving while preventing any single campaign from monopolizing the queue.

The bottleneck is verification. Models can generate possible attack paths quickly, but Apple still has to reproduce the behavior, confirm the required conditions, and decide how urgently it needs a fix. That process is inherently human. A large language model can describe a race condition in an XPC service, but only a skilled analyst can determine whether a local user can realistically win that race and what damage could follow. Automation can shorten the time to first triage, but it cannot replace the careful work of running an exploit and observing its effects.

How Apple is adapting its bug bounty program

Apple has redesigned its bug bounty program around stronger evidence. The maximum payout now exceeds five million dollars for the most serious exploit chains, while Target Flags help researchers prove that a flaw reaches protected parts of the system. Target Flags are essentially markers placed in protected areas of the OS. If a researcher's exploit can read or modify one of those markers, Apple can quickly confirm that the reported vulnerability crosses an important security boundary. That reduces the amount of back-and-forth needed to validate a submission and gives Apple more confidence that a report is real.

The higher payouts are also a response to competition. Commercial exploit brokers and state-sponsored hacking groups routinely pay more than most bug bounty programs for working zero-day exploits. Apple's increase to more than five million dollars is intended to keep researchers on the legitimate side of vulnerability research and give them a reason to report rather than sell what they find.

AI-generated vulnerability research adds another layer of complexity to that economics. On one hand, it reduces the cost of finding candidate bugs, which means more reports and more pressure on Apple's reviewers. On the other hand, it may also reduce the value of a single report because automated tools can rediscover the same issue from multiple angles. Apple's use of Target Flags and higher rewards is an attempt to reward the kind of demonstrated exploit chain that still requires considerable human skill, even in an AI-assisted world.

For Mac users, the practical takeaway is straightforward. Apple's security team is likely to miss things if it is overwhelmed by low-quality reports. The company's decision to cap open submissions and use AI for triage is an acknowledgment that the old review system was not built for this volume. The best way to stay protected is to install security updates promptly. macOS updates now include patch notes that identify the researchers who reported each vulnerability, which gives users a glimpse of how much work goes into each release.

AI bug hunting is already finding flaws that reach Apple's patch queue. The company's challenge is not finding more candidates; it is deciding which ones matter and getting fixes out before attackers can exploit the same paths. With AI making discovery easier, the future of Apple security will depend less on the quantity of reports and more on the quality of verification.

Source:Digital Trends News