Close Menu
TechurzTechurz

    Subscribe to Updates

    Get the latest creative news from FooBar about art, design and business.

    What's Hot

    Final Flash Sale: Save up to $624 on Disrupt 2025 Passes

    October 13, 2025

    I tested a Windows laptop with a tandem OLED, and it’s spoiled working on other displays for me

    October 13, 2025

    Why Unmonitored JavaScript Is Your Biggest Holiday Security Risk

    October 13, 2025
    Facebook X (Twitter) Instagram
    Trending
    • Final Flash Sale: Save up to $624 on Disrupt 2025 Passes
    • I tested a Windows laptop with a tandem OLED, and it’s spoiled working on other displays for me
    • Why Unmonitored JavaScript Is Your Biggest Holiday Security Risk
    • German state replaces Microsoft Exchange and Outlook with open-source email
    • Astaroth Banking Trojan Abuses GitHub to Remain Operational After Takedowns
    • The most important Intel Panther Lake updates are the least talked about – I’ll explain
    • Is AI even worth it for your business? 5 expert tips to help prove ROI
    • Feeling lonely at work? You’re not alone – 5 ways to boost your team’s morale
    Facebook X (Twitter) Instagram Pinterest Vimeo
    TechurzTechurz
    • Home
    • AI
    • Apps
    • News
    • Guides
    • Opinion
    • Reviews
    • Security
    • Startups
    TechurzTechurz
    Home»Apps»Agentic AI’s security risks are challenging, but the solutions are surprisingly simple
    Apps

    Agentic AI’s security risks are challenging, but the solutions are surprisingly simple

    TechurzBy TechurzAugust 19, 2025No Comments6 Mins Read
    Share Facebook Twitter Pinterest LinkedIn Tumblr Reddit Telegram Email
    AI Agent
    Share
    Facebook Twitter LinkedIn Pinterest Email


    Imagine the world’s most capable intern. Someone who can read thousands of documents overnight, make inferences from complex problems instantly, and work 24/7 without complaints. But there’s a catch: this intern is also incredibly gullible and will believe almost anything you tell them, making them the perfect target for manipulation by bad actors.

    This analogy perfectly illustrates the current state of agentic AI. It’s simultaneously the most sophisticated tool ever created and the most vulnerable to simple deception.

    This is made more challenging by how differently people view AI. The features that excite some terrify others, creating a divide between the builders and users.


    You may like

    Builders – aka engineers and researchers – focus on foundational challenges like data quality, algorithmic bias, and existential risks. Their concerns dominate headlines and academic discussions.

    But users, such as business leaders and operational teams who want to harness these tools practically and safely, have more practical worries. Less concerned about whether AI will end the world and more focused on whether it will expose customer data or make costly mistakes.

    While builders are focused on the future, users want to know what AI can do today. And unfortunately, the gap between what they expect agentic AI to deliver and what it can is substantial.

    Dr Damien Duff

    Social Links Navigation

    Dr Damien Duff is Principal AI/ML Consultant at Daemon.

    Expectation vs reality

    The narrative around agentic AI often paints a picture of fully autonomous digital workers able to alter a business overnight. While multi-agent LLMs are no longer theory, there’s more exploration that must be done before they can enable complete business transformation.

    Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!

    Current AI systems can deliver impressive agent-like behaviors including knowledge extraction from vast documents, accelerating the software delivery lifecycle, and empathetic customer interactions. But truly autonomous systems that work independently in complex novel environments remain out of reach.

    While AI can complete structured tasks with human oversight, it struggles with open-ended problems, long-term planning and high-stakes decision making where failures have consequences.

    For example, AI can identify potential vulnerabilities in code and propose wide ranging fixes, but developers must evaluate the solutions to implement and guide their application as AI can’t consistently decipher broader system context.

    The danger of inflated expectations is that organizations are caught off guard by the real security risks. Lured into a false sense of readiness, they’re ill-prepared for the routine threats they’re far more likely to face.

    The security challenge

    As agentic AI systems become more autonomous, organizations face a critical challenge of ensuring systems act aligned with business goals. As agentic AI becomes more capable, it also becomes harder to control and therefore, easier to exploit. Unlike traditional software, which fails in predictable ways, AI systems can fail creatively, manipulated in ways their creators never anticipated.

    So, what kinds of security risks should organizations actually be worried about?

    While traditional IT challenges, such as data protection across systems, risk management, robust reporting and visibility remain critical, there are other novel challenges that require fresh approaches.

    For example, adversarial prompt engineering, such as prompt injection where bad actors can embed malicious instructions in innocent requests, or context manipulation, where attackers provide false context, causing AI to make decisions based on incorrect assumptions.

    Another significant issue is accumulation of errors: while experienced humans often spot their mistakes, AI errors can quickly snowball, especially in multi-agent systems, turning small problems into big ones; this can go unnoticed in the early stages as agentic reasoning is superficially similar to human reasoning, and then quickly spiral.

    AI risks may seem daunting, but the solutions are often more familiar than organizations expect. It’s not a question of if a security problem will arise, but a matter of when, so it’s important to be prepared.

    The solution

    Many of the solutions for agentic AI’s specific security challenges exist in traditional cybersecurity and risk management frameworks. The approach requires companies to apply principles they’re already familiar with (or with which organizations they work with are familiar), including zero trust, human oversight, and controlled access.

    Rather than assuming all AI inputs are safe, companies should treat them as potentially malicious and implement multiple validation layers. This approach works across all applications, whether customer service or financial operations.

    While the principles are straightforward, successful implementation requires careful planning. To build effective agentic AI security, organizations should:

    1. Start with a small blast radius: Begin with low-risk, high-value use cases where mistakes are recoverable. For example, deploy AI for document summarization before moving onto financial transactions. This builds organizational confidence and expertise.
    2. Build governance: Don’t wait for problems to emerge. Establish clear approval processes and staff training programs so employees know what to do when things go wrong.
    3. Automate the validation process: Build and continually grow comprehensive AI-based and traditional test suites that encapsulate the worst imagined case scenarios of adversarial attacks, edge cases, and so forth, so that we are not following a moving target but measuring our improvement carefully and preventing backslides.
    4. Question everything: Train users and teams integrating AI tools to question AI outputs rather than inherently trusting them. Make verification part of company culture amongst both builders and consumers by setting up peer review processes, introducing spot-checking protocols and celebrating when someone catches a mistake.
    5. Develop protection processes: Conduct regular risk assessments, audit existing implementations for vulnerabilities, implement monitoring systems, and create AI-specific incident response procedures. Set up security policies for success.
    6. Stay ahead: Ensure employees are informed about emerging threats by providing relevant training, as well as encouraging them to participate in industry forums, discuss experiences with peers, and attend security conferences. Remember, knowledge is power.

    The future

    Agentic AI is a powerful tool that can transform how we work and solve problems. But like any powerful and emerging technology, it requires respect, understanding, and proper safety measures.

    The key is approaching this technology with the same careful planning and risk management we should apply to any other tool. The gullible savant intern analogy reminds us that even the most capable systems need supervision, clear boundaries, and ongoing guidance.

    We’ve featured the best AI website builder.

    This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

    agentic AIs Challenging risks Security simple Solutions Surprisingly
    Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
    Previous ArticleCPG startup Keychain snags $30M to build in India, grow in the U.S.
    Next Article The Download: Clean energy progress, and OpenAI’s trilemma
    Techurz
    • Website

    Related Posts

    Security

    Why Unmonitored JavaScript Is Your Biggest Holiday Security Risk

    October 13, 2025
    Security

    Hackers Access SonicWall Cloud Firewall Backups, Spark Urgent Security Checks

    October 10, 2025
    Opinion

    Reflection AI raises $2B to be America’s open frontier AI lab, challenging DeepSeek

    October 9, 2025
    Add A Comment
    Leave A Reply Cancel Reply

    Top Posts

    The Reason Murderbot’s Tone Feels Off

    May 14, 20259 Views

    Start Saving Now: An iPhone 17 Pro Price Hike Is Likely, Says New Report

    August 17, 20258 Views

    CNET’s Daily Tariff Price Tracker: I’m Keeping Tabs on Changes as Trump’s Trade Policies Shift

    May 27, 20258 Views
    Stay In Touch
    • Facebook
    • YouTube
    • TikTok
    • WhatsApp
    • Twitter
    • Instagram
    Latest Reviews

    Subscribe to Updates

    Get the latest tech news from FooBar about tech, design and biz.

    Most Popular

    The Reason Murderbot’s Tone Feels Off

    May 14, 20259 Views

    Start Saving Now: An iPhone 17 Pro Price Hike Is Likely, Says New Report

    August 17, 20258 Views

    CNET’s Daily Tariff Price Tracker: I’m Keeping Tabs on Changes as Trump’s Trade Policies Shift

    May 27, 20258 Views
    Our Picks

    Final Flash Sale: Save up to $624 on Disrupt 2025 Passes

    October 13, 2025

    I tested a Windows laptop with a tandem OLED, and it’s spoiled working on other displays for me

    October 13, 2025

    Why Unmonitored JavaScript Is Your Biggest Holiday Security Risk

    October 13, 2025

    Subscribe to Updates

    Get the latest creative news from FooBar about art, design and business.

    Facebook X (Twitter) Instagram Pinterest
    • About Us
    • Contact Us
    • Privacy Policy
    • Terms and Conditions
    • Disclaimer
    © 2025 techurz. Designed by Pro.

    Type above and press Enter to search. Press Esc to cancel.