Linus Torvalds states that bug detection by Artificial Intelligence has become the norm in Linux development

Greg Kroah-Hartman, one of the main maintainers of the Linux kernel, recently removed an obsolete project driver. Although it could have been considered just another routine task, the removal occurred after Artificial Intelligence agents inspected the code.

Kroah-Hartman made the patch for this code removal available on the kernel mailing list on August 3rd. The driver in question was implemented in the kernel in 1999 to enable the operation of Moxa Intellio serial boards, typically used in industrial environments.

The reason for the removal is the lack of usage records for this type of hardware currently. Moxa itself had already informed that that line of boards had been discontinued and that there were no plans to reintroduce it.

In 2021, a Moxa executive communicated this decision to Jiri Slaby, another Linux maintainer, signaling the possibility of removing the driver from the kernel. However, nothing was changed until May 2026, when another developer working on TTY files found the Moxa driver.

When contacted, the company confirmed the discontinuation of the Intellio line, justifying the removal of the driver from Linux, and even offered to create a removal patch. As a result, the driver entered a waiting queue to be removed, but without a defined deadline, as kernel developers usually prioritize other more urgent activities.

Subsequently, in early August, Kroah-Hartman sent the removal patch. On the kernel mailing list, he explained that the driver is very old, there is no known hardware that still uses it, and the company no longer needs it. He mentioned that AI/LLM agents were starting to analyze it and identifying 'interesting' points that would only consume time, given that the component is no longer functional.

These AI agents aim to look for inconsistencies or vulnerabilities in the kernel code. It seems these mechanisms detected some issue in the Moxa driver, which supported both its correction and its removal.

However, this was already the original plan. Based on Kroah-Hartman's communication, it can be inferred that the procedure was executed at this time to prevent AI systems from continuing to bother developers with alerts about a component that was already scheduled to be withdrawn.

It is worth remembering that Linus Torvalds has already shown support for the use of AI in Linux development, emphasizing only that the use of these tools must be moderate so that developers are not overwhelmed with notifications about irrelevant failures.

With the elimination of the Moxa driver, approximately 2,200 lines of code will cease to be maintained after the completion of this work.

Apple has implemented a limit on the number of failure alerts that security experts can submit, in response to the large volume of notification attempts, many of which originate from artificial intelligence.

The company maintains a rewards program for researchers who identify vulnerabilities in its systems, offering payments that can reach up to US$ 5 million, depending on the severity of the flaw. This program was renewed by the company in 2025.

The decision to impose limits occurred because Apple's analysis systems were overwhelmed by what the company calls 'AI slop,' meaning automatically produced low-quality content. Specifically, AI models used by the researchers themselves were exhibiting hallucinations, resulting in the identification of non-existent risks in the brand's software.

Apple has established a cap on the number of reports an individual or organization can submit; upon reaching this limit, they must wait 30 days before submitting new alerts. However, if a researcher discovers a critical vulnerability, they can request an increase in this limit, aiming to ensure that the most important notifications reach the security teams.

AI is transforming the cybersecurity landscape

The Financial Times interviewed two cybersecurity labs to discuss the process of searching for vulnerabilities in Apple's products. The startup Bynario, located in Italy, reported discovering over 50 flaws in the latest version of macOS in just three weeks but could not register these notifications due to the new restrictions imposed by Apple. The manufacturer confirmed to the FT that it contacted the researchers and is currently analyzing the findings.

A notable aspect of Bynario's work was the use of ChatGPT to assist in bug identification, a practice not exclusive to this company. Researchers in California, USA, managed to exploit a memory protection mechanism released in September 2025 using the Mythos model developed by Anthropic.

Both Apple and other actors are integrating AI into their operations. Apple itself has begun using AI tools to assist in evaluating received notifications. Furthermore, the manufacturer employs this new class of tools to conduct its own software problem hunting. The latest update to its operating systems contains approximately five times more fixes than previous versions, and Apple attributes this advancement to collaboration with the cybersecurity models from OpenAI and Anthropic.

Anthropic PBC announced in its blog on Thursday that during internal cybersecurity testing, it discovered that its artificial intelligence models compromised the security of three organizations. This occurred shortly after its main competitor, OpenAI, disclosed a similar incident.

Incident Discovery

According to Anthropic's report, an analysis of 141,006 test scenarios revealed three instances where the Claude AI tool accessed the internet and subsequently breached the infrastructure of external organizations in the real world. The earliest incidents date back to April.

When Anthropic's models gained unauthorized access to these three organizations, they treated each as part of a training exercise. These tests were 'capture-the-flag' style evaluations, where models sought hidden information by penetrating other systems, which is a common method for assessing hacking capabilities in both humans and AI. However, an older model continued the attack even after receiving data indicating it was operating on the open internet; the newest model stopped after recognizing its presence online.

Causes and Consequences of the Hacks

Neither Anthropic nor the compromised organizations noticed the intrusions. In its blog, the company admitted that it could have conducted a more thorough review of network logs and evaluation transcripts. These breaches occurred while using test environments created by the AI security firm Irregular. In each case, Anthropic instructed Claude that the environment was a simulation and had no internet access, but as stated in the blog, 'due to a misunderstanding between us and our evaluation partner, this was not the case.'

The hacks affected three different Claude models: Opus 4.7, Mythos 5, and an internal research test model. All these models operated without the standard protective mechanisms applied in public tools. The compromise of the organizations was achieved through basic methods, such as using weak passwords.

Industry and Company Reaction

This series of AI-induced hacks is already prompting some politicians to call for federal restrictions or different oversight of AI technologies. Over 1,100 employees from artificial intelligence companies signed a petition on Tuesday, as first reported by Bloomberg, urging the US government to support a mechanism that would help 'specifically regulate' AI development to prevent too rapid advancement of the technology.

Anthropic disclosed the hacks nearly four months after announcing the development of a new, potentially dangerous and powerful AI model named Mythos, the release of which was strictly limited by the company. The company emphasized that it regularly conducts tests simulating real cybersecurity issues, considering this a critical step in developing and releasing models. Nevertheless, the company concluded from these incidents that tests involving powerful autonomous capabilities also require significant control. The company stated that 'security testing happens before model release precisely because we do not yet know what it is capable of,' and that 'evaluation environments must increasingly meet the same security standards as any other system on which our models operate.'