Massive language fashions (LLMs) have turn into important instruments for organizations, with open weight fashions offering further management and suppleness for customizing fashions to their particular use circumstances. Final 12 months, OpenAI launched its gpt-oss sequence, together with commonplace and, shortly after, safeguard variants, targeted on security classification duties. We determined to guage their uncooked safety posture in opposition to adversarial inputs—particularly, immediate injection and jailbreak methods that use procedures reminiscent of context manipulation, and encoding to bypass security guardrails and elicit prohibited content material. We evaluated 4 gpt-oss configurations in a black-box setting: the 20b and 120b commonplace fashions together with the safeguard 20b and 120b counterparts.
Our testing revealed two essential findings: safeguard variants present inconsistent safety enhancements over commonplace fashions, whereas mannequin measurement emerges because the stronger determinant of baseline assault resilience. OpenAI said of their gpt-oss-safeguard launch weblog that “security classifiers, which distinguish secure from unsafe content material in a specific threat space, have lengthy been a major layer of protection for our personal and different massive language fashions.” The corporate developed and deployed a “Security Reasoner” in gpt-oss-safeguard that classifies mannequin outputs and determines how greatest to reply.
Do observe: these evaluations targeted completely on base fashions solely, with out application-level protections, customized prompts, output filtering, fee limiting, or different manufacturing safeguards. Consequently, the findings mirror model-level habits and function a baseline. Actual-world deployments with layered safety controls usually obtain a decrease threat publicity.
Evaluating gpt-oss mannequin safety
Our testing included each single-turn prompt-based assaults and extra advanced multi-turn interactions designed to discover iterative refinement methods. We tracked assault success charges (ASR) throughout a variety of methods, subtechniques, and procedures aligned with the Cisco AI Safety & Security Taxonomy.
The outcomes reveal a nuanced image: bigger fashions show stronger inherent resilience, with the gpt-oss-120b commonplace variant attaining the bottom total ASR. We discovered that gpt-oss-safeguard mechanisms present combined advantages in single-turn eventualities and do little to deal with the dominant risk: multi-turn assaults.
Comparative vulnerability evaluation (Determine 1, under) point out total assault success charges throughout the 4 gpt-oss fashions. Our key observations embody:
The 120b commonplace mannequin outperforms others in single-turn resistance;
gpt-oss-safeguard variants generally introduce exploitable complexity, that means rising vulnerability in sure assault eventualities in comparison with commonplace fashions; and
Multi-turn eventualities trigger dramatic ASR will increase (5x–8.5x), highlighting context-building as a essential weak point.
Determine 1. General Assault Success fee by mannequin grouped by commonplace vs. safeguard fashions
Key findings
Multi-turn assaults stay the first failure mode throughout all variants, with success charges climbing sharply when an adversarial attacker can refine prompts over a number of exchanges. Determine 2 under showcases the assault success fee disparities between single- and mulit-turn prompting. Particular will increase throughout the mannequin variants we examined embody:
gpt-oss-120b: 7.24% → 61.22% (8.5x)
gpt-oss-20b: 14.17% → 79.59% (5.6x)
gpt-oss-safeguard-120b: 12.33% → 78.57% (6.4x)
gpt-oss-safeguard-20b: 17.55% → 91.84% (5.2x)
Determine 2. Comparative vulnerability evaluation exhibiting assault success charges throughout examined fashions for each single-turn and multi-turn eventualities.
The particular areas the place fashions constantly lack resistance in opposition to our testing procedures embody exploit encoding, context manipulation, and procedural variety. Determine 3 under highlights the highest 10 simplest assault procedures in opposition to these fashions:
Determine 3. High 10 assault procedures grouped by mannequin
Procedural breakdown signifies that bigger (120b) fashions are inclined to carry out higher throughout classes, although sure encoding and context-related strategies retain effectiveness even in opposition to gpt-oss-safeguard variations. General, mannequin scale seems to contribute extra to single-turn robustness than the added safeguard tuning in these assessments.
Determine 4. Heatmap of assault success by sub-technique and mannequin
These findings underscore that no single mannequin variant gives enough standalone safety, particularly in conversational use circumstances.
As said at first of this publish, the gpt-oss-safeguard fashions aren’t meant to be used in chat settings. Quite, these fashions are meant for security use circumstances like LLM input-output filtering, on-line content material labeling, and offline labeling for belief and security use circumstances. OpenAI recommends utilizing the unique gpt-oss fashions for chat or different interactive use circumstances.
Nonetheless, as open-weight fashions, each gpt-oss and gpt-oss-safeguard variants may be freely deployed in any configuration, together with chat interfaces. Malicious actors can obtain these fashions, fine-tune them to take away security refusals fully, or deploy them in conversational functions no matter OpenAI’s suggestions. In contrast to API-based fashions the place OpenAI maintains management and might implement mitigations or revoke entry, open-weight releases require intentional inclusion of further security mechanisms and guardrails.
We evaluated the gpt-oss-safeguard fashions in conversational assault eventualities as a result of anybody can deploy them this fashion, regardless of not being their meant use case. The outcomes we noticed from our evaluation mirror the elemental safety problem posed by open-weight mannequin releases the place end-use can’t be managed or monitored.
Suggestions for safe deployment
As we said in our prior evaluation of open-weight fashions, mannequin choice alone can not present enough safety, and that base fashions which might be fine-tuned with security in thoughts nonetheless require layered defensive controls to guard in opposition to decided adversaries who can iteratively refine assaults or exploit open-weight accessibility.
That is exactly the problem that Cisco AI Protection was constructed to deal with. AI Protection gives the excellent, multi-layered safety that fashionable LLM deployments require. By combining superior mannequin and software vulnerability identification, like these utilized in our analysis, and runtime content material filtering, AI Protection gives mannequin agnostic safety from provide chain to improvement to deployment.
Organizations deploying gpt-oss ought to undertake a defense-in-depth technique slightly than counting on mannequin selection alone:
Mannequin choice: When evaluating open-weight fashions, prioritize each mannequin measurement and the lab’s alignment method. Our earlier analysis throughout eight open-weight fashions confirmed that alignment methods considerably affect safety: fashions with stronger built-in security protocols show extra balanced single- and multi-turn resistance, whereas capability-focused fashions present wider vulnerability gaps. For gpt-ossgpt-oss particularly, the 120b commonplace variant affords stronger single-turn resilience, however no open-weight mannequin, no matter measurement or alignment tuning, gives enough multi-turn safety with out the implementation of further controls.
Layered protections: Implement real-time dialog monitoring, context evaluation, content material filtering for identified high-risk procedures, fee limiting, and anomaly detection.
Menace-specific mitigations: Prioritize detection of prime assault procedures (e.g., encoding methods, iterative refinement) and high-risk sub-techniques.
Steady analysis: Conduct common red-teaming, monitor rising methods, and incorporate mannequin updates.
Safety groups ought to view LLM deployment as an ongoing safety problem requiring steady analysis, monitoring, and adaptation. By understanding the precise vulnerabilities of their chosen fashions and implementing applicable protection methods, organizations can considerably cut back their threat publicity whereas nonetheless leveraging the highly effective capabilities that fashionable LLMs present.
Conclusion
Our complete safety evaluation of gpt-oss fashions reveals a fancy safety panorama formed by each mannequin design and deployment realities. Whereas the gpt-oss-safeguard variants have been particularly engineered for policy-based content material classification slightly than conversational jailbreak resistance, their open-weight nature means they are often deployed in chat settings no matter design intent.
As organizations proceed to undertake LLMs for essential functions, these findings underscore the significance of complete safety analysis and multi-layered protection methods. The safety posture of an LLM shouldn’t be decided by a single issue. Mannequin measurement, security mechanisms, and deployment structure all play appreciable roles in how a mannequin performs. Organizations ought to use these findings to tell their safety structure selections, recognizing that model-level safety is only one element of a complete protection technique.
Last Observe on Interpretation:
The findings on this evaluation characterize the safety posture of base fashions examined in isolation. When these fashions are deployed inside functions with correct safety controls—together with enter validation, output filtering, fee limiting, and monitoring—the precise assault success charges are prone to be considerably decrease than these reported right here.
