Historically, organizations worried about malicious insiders, compromised credentials, ransomware operators, and careless employees as threats to their data. Artificial intelligence (AI) introduces a new concern: productive individuals with good intentions whose AI agents unknowingly behave like bad actors.
Protecting against this threat introduces a new requirement for backup applications. They must do more than protect against malicious activity by known types of bad actors. They must also offer DevOps Backup to protect organizations from legitimate, authenticated activities that unknowingly perform actions detrimental to organizational data.
Some DevOps backup providers have recognized this threat and responded with capabilities that help address it. HYCU R-Cloud for Azure DevOps represents one option for protecting Azure DevOps’ five distinct services. Organizations may use it to protect the delivery system as a whole and recover exactly what broke.
AI Can Turn ‘Good’ DevOps into Unintentional Bad Actors
An authenticated individual with legitimate permissions, such as an employee, can delegate those permissions to an AI agent. Developers represent one of the first places organizations will encounter this problem.
Developers increasingly use copilots and AI agents to perform their day-to-day tasks. These tasks include writing code, modifying applications, automating workflows, refactoring repositories, and performing work without continuous human supervision.
These individuals generally neither desire to act maliciously nor perform destructive actions. However, AI agents acting with an individual’s legitimate credentials and permissions may unintentionally behave badly.
Before AI, developers could already damage repositories through erroneous commands, force-pushes, history rewrites, misconfigured automation, and other mistakes. AI does not necessarily introduce entirely new errors. Rather, it changes their speed, frequency, scale, and autonomy.
For example, a cleanup agent might delete branches it incorrectly considers stale. A refactoring agent might force-push after a bad rebase. A migration script might rewrite multiple repositories, while an AI-generated CI workflow might execute a destructive Git command.
Previously, developers manually executing these commands made performing root cause analyses more manageable. AI may now execute tens, hundreds, or thousands of commands in seconds. These actions appear legitimate because they occur using authenticated and authorized credentials.
Organizations may not discover their impact for some time. Even the ‘good’ individual who authorized the AI agent may not know anything went wrong.
Git Repositories Now Contain Much More Than Source Code
Aggravating the situation, GitHub, GitLab, Bitbucket, and Azure DevOps increasingly contain or connect more than application source code. They may include infrastructure-as-code, deployment configurations, runbooks, security policies, AI prompts, and agent definitions.
In yesterday’s pre-AI world, losing a repository primarily risked losing source code and associated development data. Today, losing or corrupting one may mean losing part of an organization’s operating model.
That risk becomes even more pronounced with Azure DevOps. It includes Repos, Boards, Pipelines, Test Plans, and Artifacts, each possessing different data models and relationships. Simply restoring files may therefore fail to restore a usable development environment.
This creates a broader concern than restoring repositories that authorized AI agents inadvertently change or delete. As organizations adopt AI-driven internal workflows, DevOps platforms may increasingly contain instructions defining how applications and potentially the business itself operate.
AI Changes the Value of 30 Minutes of Lost Data
Developers using authorized AI agents introduce another variable for which DevOps backup must account: time. AI agents can accomplish in hours or minutes what developers previously needed days or weeks to complete.
Agents and copilots may also work overnight and through weekends without continuous human supervision. This accelerated development makes data stored in DevOps platforms more time-sensitive and increases how much one destructive event might erase.
Historically, organizations might have considered daily or 12-hour Git repository DevOps backup sufficient. AI agents dramatically increase development velocity, making those recovery point objectives (RPOs) increasingly risky.
Organizations should consequently re-evaluate their existing RPOs. Thirty minutes of lost AI-generated work may now represent what previously required developers hours or even days to recreate.
DevOps Backup Must Protect Known-Good Data from Authorized Activity
Traditional cyber-resilient DevOps backup often assumes an attacker first compromises production before attempting to compromise or delete backups. AI complicates this model because an AI agent may legitimately inherit an employee’s credentials, permissions, tokens, and API access. The production system may now have no reason to reject the AI agent’s actions. This creates two new DevOps backup requirements.
- First, organizations need independent recovery points that allow them to return applications or DevOps data to known-good states.
- Second, DevOps backup must remain beyond the reach of identities, tokens, APIs, and AI agents operating against production.
Immutability consequently assumes a broader role. It continues protecting backups from ransomware operators and malicious administrators. However, it also creates a boundary between autonomous AI activity and an organization’s known good recovery copies.
HYCU Expands DevOps Backup from Repositories to the Development System
Organizations facing these issues can look to HYCU for one approach to addressing them. HYCU already protects GitHub, GitLab, and Bitbucket. Adding Azure DevOps extends its DevOps backup approach beyond repositories toward protecting the broader software development lifecycle.
HYCU R-Cloud for Azure DevOps protects work-item histories, boards, queries, pipeline definitions, wikis, test plans, organization settings, process templates, and custom work-item types. Organizations can use it to recover individual items, repositories, pipelines, projects, or organization settings.
HYCU also combines Object Lock and WORM immutability with customer-owned storage. Organizations can keep protected copies outside Azure DevOps’ native recovery mechanisms and establish retention policies independently from those mechanisms.
Further, HYCU keeps backup and recovery logically separate from the applications it protects. Recovery need not depend upon the same application, identity, token, or automated workflow that may have caused the damaging change. HYCU also supports backup intervals as frequent as 30 minutes, which aligns the recovery point with the pace at which agents now generate work.
This architecture illustrates a broader requirement emerging with agentic computing. Backup must provide an independent recovery layer that remains available when legitimate, authenticated AI-driven activities modify, corrupt, or delete production data.
The Next Bad Actor May Be Your Best Employee
The best organizations want their developers to work more efficiently and effectively. They want developers using AI because it can help them accomplish substantially more work much faster.
However, granting AI access also introduces organizational risks. DevOps backup must account for the consequences of granting AI agents the permissions needed to deliver these productivity gains.
HYCU R-Cloud addresses these risks by creating independent, immutable backup copies of DevOps data. It protects GitHub, GitLab, Bitbucket, and Azure DevOps while providing granular recovery of known-good data when AI agents unintentionally modify, corrupt, or delete it.
Organizations may therefore discover their next bad actor possesses no malicious intentions or stolen credentials. It may represent one of their best employees using AI exactly as encouraged. HYCU’s independent, immutable recovery copies let organizations embrace AI’s productivity without making their employees’ unintended AI mistakes permanent.
KEEP UP TO DATE WITH DCIG
To be notified of new DCIG articles, reports, and webinars, sign up for DCIG’s free weekly Newsletter.
To learn about DCIG’s future research and publications, see the DCIG Editorial Calendar.
Technology providers interested in licensing DCIG TOP 5 reports or having DCIG produce custom reports on their behalf, please contact DCIG for more information.
Editor’s note: HYCU is a client of DCIG