Evaluating Jev: Can the New Encoder-Based Model Serve as an Effective Guard for LLMs?
The cybersecurity community continues to explore specialized guard models that can protect large language models from jailbreaks, prompt injection, and data exposure. A new analysis from Raft asks whether Jev, released by TypeSafe on 15 September, can serve as a practical guard model for both safety and privacy use cases.
Over the past two years, open-source guard models have largely been dominated by large language models such as Qwen Guard, gpt-oss-safeguard, and YuFeng XGuard. Earlier encoder-based attempts, including Wildguard built on Mistral 7B, offered strong classification quality but suffered from slow inference and high memory requirements. Most BERT-style encoders lacked multilingual pre-training, making them ineffective for Russian text, while LLMs inherited multilingual capabilities from their pre-training.
In May 2026, three new encoder-based guard models appeared almost simultaneously: gliner guard, gliguard from Fastino, and the gliclass family. These models represent a shift back toward efficient encoder architectures. Jev distinguishes itself by offering a unified model for both safety and privacy detection and by claiming an unusually large context window of 32K–64K tokens, far exceeding the typical 512–8192 token limits of most encoders.
Testing shows that Jev handles obvious jailbreaks and prompt-injection attempts effectively. It can indicate whether a text contains personally identifiable information, yet it does not perform classic named-entity recognition and cannot return exact text spans for masking. Consequently, production deployments would require an additional NER stage or full request blocking.
On the Aegis 2.0 benchmark from NVIDIA, Jev achieved an F1 score of 0.85 and recall of 0.84 on prompts at the 0.5 threshold, with an AUC of 0.921. On model responses the AUC rose to 0.927, although recall dropped to 0.76. In the broader GuardRate evaluation, Jev placed second with a score of 0.835, behind only the 0.4B opir-multitask-large GliClass model (0.906) and ahead of Qwen3Guard-Gen-8B (0.819) and both YuFeng XGuard variants.
Despite these promising results, two practical blockers remain for production use: all inference currently routes through an external API, raising data-protection concerns under Russian Federal Law 152-FZ, and the model has not yet been evaluated on Russian-language traffic. Organizations considering Jev are advised to validate domain-specific performance and false-positive rates on their own datasets before deployment.
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.