Forgotten AI agents may still access sensitive company data

Help Net Security

October 26, 2026 at 5:30 AM

Overview

A recent report from Arrakis Security reveals a concerning issue with AI agents in corporate environments. It found that about 76% of these agents had not been active for over 60 days, yet they still maintained access to sensitive company data through retained credentials and connections. This situation poses a significant risk, as these inactive agents can perform various actions, including reading and writing data or initiating workflows. Companies that utilize AI agents need to be aware of this vulnerability and take steps to manage and disable agents that are no longer in use, to prevent potential unauthorized access to their systems and data. The implications are serious, especially for organizations handling sensitive information.

Key Takeaways

Affected Systems: AI agents in corporate environments, company data systems
Action Required: Companies should regularly audit and disable inactive AI agents to prevent unauthorized access.
Timeline: Newly disclosed

Original Article Summary

AI agents that have not run recently retain credentials and connections to company systems, according to Arrakis Security’s 2026 Autonomous Agents Threat Report. About 76% of enabled agents examined had not run in 60 days, leaving access available through inactive software. Agents by exposure family (Source: Arrakis Security) The analysis covered thousands of agents in customer production environments over a 12-week period. These programs can read and write data, call tools and start workflows, giving … More → The post Forgotten AI agents may still access sensitive company data appeared first on Help Net Security.

Impact

AI agents in corporate environments, company data systems

Exploitation Status

The exploitation status is currently unknown. Monitor vendor advisories and security bulletins for updates.

Timeline

Newly disclosed

Remediation

Companies should regularly audit and disable inactive AI agents to prevent unauthorized access.

Additional Information

This threat intelligence is aggregated from trusted cybersecurity sources. For the most up-to-date information, technical details, and official vendor guidance, please refer to the original article linked below.

Related Topics: This incident relates to Vulnerability.

Read Original Article

Related Coverage

Synthetic identities leave US lenders with $2.4 billion in fraud exposure

Help Net Security

Criminals are increasingly using synthetic identities to exploit established customer accounts, resulting in an estimated $2.4 billion in potential fraud exposure for U.S. lenders, according to a report by TransUnion. During the first half of 2026, a staggering 81% of suspected digital fraud incidents occurred during account logins and financial transactions. This indicates that simply verifying a person's identity at the start of a relationship does not guarantee trust in future interactions. Scammers are adept at manipulating legitimate customers, making it crucial for companies to enhance their fraud detection and prevention measures to protect sensitive information and financial assets. The report serves as a wake-up call for financial institutions to reassess their security protocols in light of these evolving tactics.

Oct 26, 2026

OpenSSL 4.1.0: What’s new, what’s removed, and what’s still broken

Help Net Security

The OpenSSL project has released version 4.1.0, which introduces support for DTLS 1.3, a protocol variant designed for datagram transport over networks like UDP. However, this new version comes with four known bugs in the DTLS code, which could lead to applications misinterpreting connection data. Developers using DTLS 1.3 or the recently added DTLS support in OpenSSL’s SSL listener API are particularly at risk. This situation is concerning as it could affect the reliability and security of applications that rely on these protocols, potentially leading to data miscommunication. Users and developers should stay alert for updates or patches that address these issues.

Oct 25, 2026

Unpatched Tailscale SSH Flaws Could Let a Hacked Device Reach Root on Other Servers

The Hacker News

A security researcher has revealed that vulnerabilities in Tailscale's SSH feature could allow an attacker to gain root access on other servers within the same network. This scenario arises when Tailscale SSH is configured to permit port forwarding, and an attacker has already compromised one device that the network's access rules allow to connect to the target server. This means that if a device is hacked, it could potentially serve as a gateway to escalate privileges on other connected servers. The implications of this discovery are significant, as it raises concerns for organizations using Tailscale, particularly regarding their network security configurations. Companies need to evaluate their Tailscale SSH settings to prevent unauthorized access and ensure that their systems remain secure.

Oct 25, 2026

PHP2Shell Exploitation Hits Government Entity Amid Global Scanning and Signs of AI-Assisted Hacking

The Hacker News

Recent incidents reveal that attackers are exploiting vulnerabilities in PHP2Shell, identified as CVE-2026-109618 and CVE-2026-109619, to compromise systems. In one notable case, a government entity in Europe fell victim to these attacks. Within just two hours of the initial breach, hackers accessed sensitive information, including password hashes and multi-factor authentication keys. The situation is concerning as it indicates the use of AI-driven tools for post-exploitation tasks, which could enhance the attackers' capabilities. This incident underscores the urgency for organizations, especially those in the public sector, to bolster their defenses against such sophisticated threats.

Oct 25, 2026

AI coding agents hijacked using AGENTS.md file in Downloads folder

BleepingComputer

Researchers discovered a malicious file named AGENTS.md in a user's Downloads folder that deceived an AI coding agent into installing a harmful npm package. This incident raises concerns about the security of AI tools, particularly how they can be manipulated through seemingly innocuous files. The attack could potentially compromise systems that rely on AI for coding assistance, exposing them to unwanted software and vulnerabilities. Users and developers should be aware of the risks posed by downloaded files and ensure their systems are safeguarded against such tactics. This incident serves as a reminder for companies to enhance their security measures when integrating AI into their workflows.

Oct 25, 2026

OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication

The Hacker News

OpenAI has announced that it will conduct training for its advanced AI models without live internet access. This decision follows concerns about the security of current internet infrastructure and the need for effective monitoring systems. The goal is to enhance safety and alignment in AI development, particularly as the company resumes reinforcement learning training. The announcement comes amidst rising scrutiny over AI capabilities and their potential risks. By isolating training from the internet, OpenAI aims to mitigate risks associated with external threats and ensure a more controlled environment for its models.

Oct 25, 2026