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

Curious whether open models make AI safer or riskier — I'll survey the latest arguments for both sides.

Open-source AI is neither categorically safer nor categorically riskier than closed / proprietary AI - the two models create different types of safety and risk, and experts disagree on which risk is more serious.

1. Why open-source is argued to be safer

Transparency and auditability: Proponents argue open systems can be inspected, audited and patched by a wide community, while closed systems are black-boxes.
> "Increased transparency allows the community to audit systems for security risks, run their own versions of the models, and learn best practices for managing AI infrastructure."

By analogy to open-source software, tools such as Linux are often seen as safer than closed-source counterparts because vulnerabilities have been probed and patched by a vast community of contributors.

Similarly, researchers can more easily test open models for security flaws, and supporters argue vulnerabilities surface faster when thousands of eyes examine code and weights:

> "In a world where cybersecurity attackers use advanced AI, defenders need access to models with comparable capabilities so they can detect, simulate, and respond to emerging threats"

That includes cyber-defence: open models can be deployed more broadly than closed models for cyber defense, because any tool that can be used for hacking can also be used to prevent it. After Hugging Face was hacked by OpenAI agents, it had to depend on an open model to analyze the hack because questions about it triggered closed models' security filters.

Prevents concentration of power: The clearest argument for open models is around preventing mass concentration of power — making sure foundational technology is a public good.

A closed model can create concentrated risk inside a small number of extremely powerful companies, while an open model distributes capability. Without open options, a handful of labs decide who accesses frontier intelligence and on what terms.

> "transparency inherent in open models can enhance safety and trust... This contrasts with closed systems, where lack of visibility can obscure risks and limit accountability."

Mark Zuckerberg in 2024 argued open source AI will be safer than alternatives against unintentional harm, because open systems can be widely scrutinized.

Regulatory view: The 2024 NTIA report on Dual-Use Foundation Models with Widely Available Model Weights took a cautious yet optimistic view that benefits of open source AI will outweigh risks, including promoting innovation and forestalling algorithmic monoculture, and recommended active monitoring rather than immediate restriction. It found evidence too immature to declare open models riskier on balance.

> "current evidence is not sufficient to conclude that restrictions on open-weight models are warranted now"

A U.S. Commerce Department assessment reported by AP also argued open systems are less of a risk than closed counterparts because transparency can help mitigate risks such as bias but also spur innovation.

2. Why open-source is argued to be riskier

Irreversible release and loss of controls:

The debate between open and closed is often framed as Transparency vs control: Open systems can be audited, while closed systems can be throttled and monitored by the owner.

Closed providers use automated filters to detect hacking, biological weapons or other risky queries and can shut down accounts when filters are bypassed. When someone downloads a model and runs it on their own hardware, there's no filter checking prompts and no way to revoke access.

Once weights are downloaded and run independently, any safety measures applied to the hosted version become completely unenforceable.

Easy removal of safeguards:
Companies train models to refuse harmful requests, but researchers have shown it's surprisingly easy to retrain a "safe" open model to agree to engage in previously refused behaviors. Open-source AI can be easily modified or 'jailbroken' to produce harmful content.

> "Downloaded open models live on user hardware. Once modified, they run unchecked."

Higher misuse potential for catastrophic harms:

Proponents of closed-source argue releasing powerful weights is dangerous, potentially providing tools for biological weapons or sophisticated cyberattacks.

Risks assessed include cybersecurity attacks, chemical and biological misuse, mass persuasion, and self-replicating models.

Anthropic's 2026 report on GLM-5.3, a Chinese open-weight model, warned it poses major cyberattack risk precisely because it is freely available for anyone to download and modify.

Where it gets complicated is low-probability but catastrophic incidents: if a wildly powerful open hacking model were dropped today, a bad actor could devastate infrastructure before defenses are ready.

> "many experts still perceive open models as riskier than closed models. The time-tested open-source security logic doesn’t necessarily hold up when dealing with tools as powerful as today’s—and tomorrow’s—models."

Public perception reflects this: in a McKinsey survey, respondents considered open source AI tools riskier than proprietary AI for most AI risks, citing cybersecurity (62%), regulatory compliance (54%) and IP (50%).

Critics also note transparency doesn't automatically fix other risks:

> "while open-source AI ... may provide a higher degree of transparency compared to proprietary AI technologies, other risks including hallucinations and inaccuracies, capacity to cause harm, infringement of intellectual property, and privacy risks are not necessarily mitigated"

3. Why closed AI is argued to be safer — and riskier*

Safer aspects of closed:

Vetted, enterprise-ready versions of open models — often maintained by cloud providers — are considered a safer bet than deploying raw unverified weights found on public repositories.

Riskier aspects of closed:

4. Synthesis: Trade-off, not absolute

Most recent analyses conclude:

> In practice, the AI ecosystem likely needs both approaches serving different purposes and risk profiles.

Current policy direction in the US and EU has been to allow openness while developing monitoring: NTIA recommended not restricting open weights immediately but actively monitoring risks and preparing to act if heightened risks emerge, and the EU AI Act makes exceptions for some open-sourced models developed with full transparency while researchers work on mitigations like data curation and tamper-resistant training.

Bottom line:

Many experts therefore view open source as safer for society as a whole if managed well, but riskier in terms of direct, uncontrolled misuse — and advocate mitigations rather than all-or-nothing bans.