Is open-source AI safer or riskier than closed/proprietary AI?

This is a genuinely contested question in AI policy circles, with legitimate arguments on both sides. Let me pull in some current perspectives to give you a well-rounded view.## There's no simple answer — it depends on what kind of "safety" you mean

This is one of the most contested questions in AI policy, and reasonable experts disagree substantially. The 2025 International AI Safety Report noted wide disagreement among experts about the likelihood of losing control over advanced AI systems. Here's the case each side makes:

The case that open-source/open-weight AI is safer

More eyes find more flaws. With open weights, independent experts can evaluate risks, biases, and vulnerabilities. Supporters argue transparency can improve security because researchers, developers and security experts around the world can examine them, identify vulnerabilities and design improvements, and with a broader community contributing to testing and improvement, issues may be discovered and addressed more quickly.

Closed models aren't inherently safe either. Microsoft's public position puts it bluntly: openness may be one of the most important paths to AI safety and security, since relying solely on closed models is not inherently safe — they can be breached, misused, or fail in ways that outsiders cannot detect, and concentrating advanced AI capabilities behind a small number of closed models compounds that risk, resulting in a small number of single points of failure that weakens competition and leaves critical technology in the hands of a few providers.

Defenders need the same tools attackers have. A coalition of major tech and cybersecurity firms recently argued that open weights pose "real and distinct risks" because anyone can modify the AI models after their release, creating a sprawl of potentially malicious variants, but they also "broaden defensive capability, increase transparency, and allow vulnerabilities to be discovered and remediated across many teams." They went further, calling open-weight models "one of the most important paths to AI safety and security," noting that closed-source models can be hacked or abused in hard-to-detect ways.

Avoids concentration of power. Some argue open weight AI could act as a check on large tech firms, or disperse the economic benefits of AI further.

Sovereignty and sensitive-sector use. Open-weight models empower a broader range of companies and governments to access, adapt and deploy AI locally, including in critical sectors where data is deemed too sensitive to share with third parties, such as healthcare and national security.

The case that open-source/open-weight AI is riskier

Loss of control is permanent. Once weights are out, openness is not without its dangers — once model weights are released, control over how the model is used is effectively lost, and safeguards can be circumvented. Microsoft itself concedes open weights carry real and distinct risks: once released, the weights are beyond the original developer's control, and modified versions are difficult to trace or reverse.

Safety guardrails are easy to strip out. The International AI Safety Report notes that risks posed by open-weight models are largely related to enabling malicious or misguided use, since general-purpose AI models are dual-use and open-model weights can exacerbate misuse risks by allowing actors without the resources to build a model themselves to leverage existing capabilities without oversight — and while both open and closed models can have safeguards, these are easier to remove for open models.

Lowers the barrier for bad actors. Critics worry such systems could pose safety risks by increasing the chances of bad actors using these systems for harmful purposes including learning how to create bioweapons and carry out cyberattacks. More specifically on capability-sprawl: critics argue that sufficiently capable models could be adapted to support harmful activities, including sophisticated cyberattacks or the development of chemical and biological weapons.

Even China's government, generally pro-open-source, now flags this. A recent Chinese AI safety framework acknowledged risks from models with open weights, including the downstream propagation and amplification of model defects, which could create vulnerabilities in open-source base models that could cascade through the ecosystem as they are fine-tuned and deployed by users, and warned open-source models could make it "easier for criminals to train 'malicious models.'"

New supply-chain/social-engineering risks. Beyond model misuse, open ecosystems face AI-amplified threats: deploying an open source LLM introduces risk of unauthorized access to the model or sensitive data if security controls aren't in place, and malicious actors can tamper with training data or the model itself, injecting harmful code or biases.

The honest synthesis

Most serious analyses land on "it depends" rather than a clean verdict:

Bottom line: Open-source/open-weight AI tends to be safer along dimensions like transparency, auditability, avoiding vendor lock-in/single points of failure, and giving defenders equal tools to attackers — but riskier along dimensions like irreversibility, ease of stripping safety guardrails, and lowering the skill/resource floor for malicious actors. Capability level matters too: the real flashpoint is specifically around frontier-level open releases (like Llama, DeepSeek, or GPT-OSS) that approach the danger threshold for things like bioweapons or cyberattack uplift, not open-source AI in general.