AI Agent Workflow Exceeds 100-Unit ERP Limit Reaching 120 Due to Check-Then-Act Race Condition
A new case study examines how an AI workflow authorized to order no more than 100 units of goods nevertheless created orders totaling 120 units in an ERP system. The scenario involved three separate orders of 40 units each to suppliers A, B, and C. Every individual request passed through the authorization layer and received an ALLOW response, yet the final ERP state showed 120 units.
The core issue lies in the difference between local and global authorization checks. A simple rule checking if requested_quantity <= 100 answers only whether the current order stays under the threshold. It does not verify whether the sum of all orders will remain within the original human-delegated limit of 100 units for COMP-17 across approved suppliers.
Sequential stateful checks correctly accumulated consumption and rejected the third order, leaving 80 units in the ERP. However, when requests executed concurrently, all three handlers read the initial balance of zero before any updates, allowing all three ALLOW decisions. The final ERP state reached 120 units with no lost updates or double counting.
The corrected implementation performs an atomic reservation inside a database transaction using PostgreSQL SELECT ... FOR UPDATE. This reserves the requested quantity from the grant balance before the ERP call occurs, ensuring that once 40 units are reserved they become unavailable to other concurrent requests. In concurrent tests, two orders received ALLOW while the third was rejected with DENY_CAPACITY, resulting in a correct total of 80 units.
The study emphasizes that an ALLOW decision only proves the authorization component approved the request. Without atomic reservation tied to the overall grant, multiple locally valid decisions can collectively exceed the delegated mandate. The experiment deliberately used a deterministic planner without any LLM to focus purely on the authorization boundary.
Five practical questions are recommended before deploying agents with cumulative business limits: what exactly the human authorized, how actions are linked to a single grant with scope and remaining balance, whether ALLOW is bound to atomic reservation, whether parallel branches or retries can overspend the same balance, and how final state is independently verified beyond authorization logs.
Full reproduction materials, 44 passing tests, and evidence files are published under the tag shared-cumulative-authority-v1.0 on GitHub. The findings apply to any agentic system where multiple actions consume a single finite human-delegated mandate such as purchase volumes, payment limits, or message quotas.
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.