AI

BeyondTrust predicts AI agents will reshape cybersecurity risks in 2027

BeyondTrust expects autonomous AI agents, faster vulnerability exploitation and closer scrutiny of AI governance to reshape cybersecurity in 2027, as new industry standards emerge for agent identity.

[For more news, click here.]

AI agents are gaining access to business applications, data and tools to complete tasks with limited human involvement. That creates a new security challenge: a system operating with legitimate permissions may still expose sensitive information or make harmful changes. Security teams must therefore control not only which agents connect, but what they can do.

Cybersecurity company BeyondTrust has identified autonomous agents, excessive access permissions and AI-assisted attacks as major risks in its cybersecurity predictions for 2027 and beyond. Its forecast also anticipates demands for evidence that AI controls work in practice. These are forecasts rather than confirmed incidents or regulatory decisions.

AI agents create new identity and access risks

BeyondTrust predicts that a major AI agent-related security incident could resemble an insider threat because the agent would be acting through valid credentials. Investigators might need to determine whether the system exceeded its intended role, who authorized its access and who has authority to shut it down without disrupting operations.

The company also warns that over-privileged agents could become a significant route into sensitive systems. If an attacker manipulates an agent through malicious instructions or a compromised connection, the damage may be limited less by the model itself than by the permissions already granted to it. BeyondTrust argues that restricting agents to the minimum access needed for their tasks will become central to security planning.

Industry efforts are beginning to address part of this problem. On November 10, the Linux Foundation welcomed Agent Name Service (ANS), an open standard that uses existing internet infrastructure to help systems discover and verify AI-agent identities. The project has support from 11 organizations, including Cisco and Cloudflare. Identity verification, however, is different from authorization: confirming which agent is acting does not establish which files, services or actions it should be permitted to access.

Faster attacks challenge traditional security responses

Another BeyondTrust prediction concerns AI-assisted vulnerability discovery, particularly in open-source software. The company expects automated research to increase the number of reported weaknesses and make a strategy of attempting to patch every flaw increasingly difficult. Security teams may need to prioritize vulnerabilities based on actual exposure, potential impact and threat intelligence rather than treating every issue equally.

It also raises the possibility of an end-to-end autonomous breach in 2027. In that scenario, AI agents could identify weaknesses, combine exploits and extract data with limited direct human involvement, while defenders respond using their own automated systems. BeyondTrust presents this as a potential development, not an event that has already occurred.

Some of the operational complexity is visible in the company's earlier research. In a four-week study, BeyondTrust Phantom Labs observed approximately 1.28 million shell commands requested from 132,000 AI coding-agent prompts, or nearly ten commands per prompt. It also reported that 85% of observed coding-agent users had at least one Model Context Protocol (MCP) server connection, linking agents to additional services. The figures cover its sample, not all enterprise AI deployments.

AI governance moves from policies to proof

BeyondTrust expects organizations to face closer scrutiny of AI security controls from auditors, regulators, insurers and customers. Companies may increasingly need records showing which agents acted, what permissions they held and how their activity was monitored, rather than relying on policies that are difficult to verify.

The wider forecast also considers credentials left on discarded devices, AI-generated answers, quantum computing and robotics.

For businesses deploying autonomous systems, the practical question is how to preserve the advantages of automation without allowing permissions and responsibilities to expand unchecked. BeyondTrust's forecast does not establish when a major incident will occur, but it highlights a concrete task for security teams now: inventory AI agents, limit their access and decide in advance how harmful activity can be contained.

Related Articles

Abu Dhabi Launches AI Tools to Rewrite Arabic Texts for Different Reading Levels
AI

Abu Dhabi Launches AI Tools to Rewrite Arabic Texts for Different Reading Levels

Etisalat
AI

Etisalat UAE Unveils AIO as e& Expands Its Sovereign AI Push

Saudi Arabia expands Mostaqbali AI training program with 12 new courses
AI

Saudi Arabia expands Mostaqbali AI training program with 12 new courses

Qatar launches two AI projects to support schools and university students
AI

Qatar launches two AI projects to support schools and university students

AI generated Image
AI

Etisalat UAE and Khalifa University Push AI Into the Architecture of 6G

Vodafone, Elsewedy and Cassava plan $1 billion data center expansion in Egypt
AI

Vodafone, Elsewedy and Cassava plan $1 billion data center expansion in Egypt