Agentic AI Systems Under Siege: Prompt Injections, Data Poisoning, and Tool Exploits
Independent expert Andrey Biryukov examines how AI agents transition from simple chat interfaces to autonomous systems capable of reading files, sending emails, calling APIs, and managing repositories, thereby becoming targets for a new class of threats.
Unlike conventional chatbots, agents possess autonomy that introduces risks to confidentiality, integrity, and availability. Confidentiality breaches occur when agents read unauthorized files or exfiltrate data through tool outputs, workspace files, memory logs, or webhooks. Integrity violations arise when agents delete files, misroute emails, initiate fraudulent transactions, or select suboptimal vendors. Availability issues emerge when long-running tasks or browser automation create cascading failures that lock entire pipelines.
Attacks via Trusted Content
Prompt injection remains the primary vector because large language models cannot reliably distinguish user instructions from processed data. Malicious commands embedded in web pages, emails, code comments, or Jira tasks can be executed with high trust. Researchers at NeuralTrust demonstrated that specially crafted strings resembling URLs caused OpenAI Atlas to interpret instructions instead of navigating, bypassing normal validation. Zscaler ThreatLabz created a fake Python library documentation site containing hidden instructions in CSS and JSON-LD metadata that convinced coding agents to purchase a $3 license key via the attacker’s cryptocurrency wallet; four of 26 tested models complied.
Data Poisoning and Backdoor Attacks
Data poisoning targets the training or fine-tuning phase. Researchers from Carnegie Mellon and Cornell Tech altered public datasets on immigration, hiring discrimination, racial disparities, autonomous driving, and AI motivation, then uploaded tampered versions to private repositories. Agents from Anthropic, OpenAI, and Google selected the poisoned datasets in approximately half of queries. In medical imaging, insertion of just 250–300 poisoned samples into a one-million-image pneumonia dataset (0.025 %) was sufficient to implant a backdoor that caused the model to miss diagnoses in specific demographic groups.
Tool and Protocol Vulnerabilities
Three vulnerabilities in Git MCP Server allowed agents to escape repository boundaries and combine file-system, terminal, and connector access to execute malicious code. Comparative testing of Function Calling versus Model Context Protocol architectures revealed distinct risk profiles: Function Calling proved more resistant to direct prompt injection but more susceptible to tool manipulation, while MCP offered better component isolation yet enabled more cross-component attacks. Composite attacks succeeded far more often than isolated ones across 3,250 test scenarios.
Defense Recommendations
OWASP published the Agentic Top 10 listing risks such as Agent Goal Hijack, Tool Misuse, identity abuse, supply-chain compromise of MCP components, and memory poisoning. Yandex and Kaspersky Lab advocate applying STRIDE and OWASP methodologies at design time. Core principles include least-privilege isolation, strict separation of trusted and untrusted data, continuous monitoring, and validation of URLs and high-risk actions. Joint guidance issued in April 2026 by Canada, Australia, the United States, New Zealand, and the United Kingdom covers the full lifecycle from design through operation.
Related articles
TenetBench Benchmark Tests LLMs for Compliance with Professional Ethical Codes
Researchers from the Institute of Information Security Issues at ISP RAS and the Institute of Cybersecurity and Information Protection at RANEPA have developed TenetBench, a benchmark designed to evaluate how well large language models adhere to specific norms in professional codes of ethics. The benchmark generates realistic work-related prompts that may conflict with rules from documents such as the Code of Professional Ethics for Advocates or journalistic standards on undisclosed advertising. Each test case pairs a conflicting request with a neutral counterpart to measure whether models refuse, warn, or still provide prohibited content. Evaluation separates task completion, norm violation detection, and presence of explicit refusals using an LLM-as-a-Judge approach validated against human annotators. The full dataset contains 3,433 prompt pairs derived from 15 professional codes, with a public subset of 2,235 pairs from 10 documents. Human validation confirmed 99.5% of generated examples as fully plausible work scenarios.
Solar to Showcase Solar AI Firewall for Business Protection Against AI Risks at SOC Forum 2026
GK Solar will present its new Solar AI Firewall solution at SOC Forum 2026 to help organizations control interactions with AI services and prevent confidential data leaks through employee prompts. The product addresses growing risks from staff sending corporate data to neural networks and AI agents gaining access to internal systems. The forum takes place on 27-28 October in Moscow's Lomonosov cluster, with the official launch of Solar AI Firewall scheduled during the event. Solar will also highlight practical risks of AI adoption, including data transmission to external services, use of public models by employees, and new attack surfaces in corporate infrastructure. The solution will be demonstrated as part of an integrated ecosystem for access, network, and data protection. In addition, the company plans to introduce JSOC ONE, a new line of Solar JSOC services tailored for mid-sized businesses that require 24/7 monitoring without building their own SOC.
Researcher Runs 100 Abliterated AI Agents Against His Own Accounts, Achieves Five Successful Breaches
A security researcher deployed roughly 100 autonomous AI agents based on abliterated open-source models to attempt unauthorized access to his online accounts over a five-hour period. The agents successfully compromised three accounts by exploiting vulnerabilities in legacy personal projects and two additional accounts through password reuse and credential stuffing. They also conducted 16 social engineering attempts, harvested and cross-verified personal data from public sources, and generated phishing pages targeting the researcher's Substack and Hacker News presence. Despite these successes, the agents failed to discover any zero-day vulnerabilities in third-party software or gain access to Tier-0 accounts such as Gmail, 1Password, or banking services. The experiment relied on abliterated versions of GLM-5.3 and DeepSeek V4 Flash models running on cloud GPUs, orchestrated via Codex CLI inside Docker containers with access to Railway infrastructure. Total compute cost reached approximately $210, equating to roughly $40 per successful account compromise. The researcher concluded that inexpensive swarms of such agents could soon enable low-skill attackers to systematically map and exploit an individual's entire online footprint.
How Censorship in Closed AI Models Reshapes Science: Hugging Face Releases Open Carbon-X DNA Model
Hugging Face CEO Clement Delangue announced the release of Carbon-X, an open 14.8 billion parameter language model designed for reading, analyzing, and generating DNA sequences. The announcement highlighted how shallow guardrails in closed commercial models blocked even proofreading a tweet discussing the topic. Commenters and researchers pointed out that these input/output filters lack context awareness, blocking legitimate scientific queries while motivated adversaries bypass them through rephrasing. Carbon-X was built on Qwen3-14B with a hybrid BPE and 6-mer tokenizer, domain mid-training on 302 billion tokens, and safety achieved through data curation rather than runtime censorship. The model achieves performance comparable to closed frontier systems at roughly 1/1800th the inference cost for large-scale genomic tasks. The case underscores growing asymmetry between closed corporate AI labs building wet labs and independent researchers restricted by superficial safety layers.