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

Open-source vs closed AI is a contested safety question — I'll pull the evidence for both sides.

Is open-source AI safer or riskier than closed/proprietary AI? The answer is: neither is strictly safer — they have different risk profiles and trade-offs.

1. The Core Debate

The industry is at a crossroads between two philosophies:

Experts note that neither approach offers a perfect solution.

2. Arguments That Open-Source Is Safer

a) Transparency and accountability: Open models provide visibility of algorithms and data handling, which leads to better accountability. That transparency is seen as critical in sensitive applications like healthcare or criminal justice.

Proponents argue "security through obscurity" is a fallacy, and open-source is actually safer because it eliminates single points of failure and prevents a small group of engineers from having unilateral control.

Similar to Linux, open-source tools are often seen as safer because vulnerabilities have been probed and patched by a vast community of contributors.

b) Wider defensive use: Any tool that can be used for hacking can also be used to prevent it. Open models can be deployed much more widely for cyber defense than closed models limited to trusted partners via Project Glasswing-style programs. After it was hacked by OpenAI agents, Hugging Face emphasized it had to depend on an open model to analyze the hack, because questions triggered closed models' security filters.

c) Decentralization of power: The clearest argument for open models is preventing mass concentration of power — making foundational technology a public good. If frontier systems are controlled entirely by select people, companies or governments, those groups have increasing control over what the world looks like.

d) Regulatory view: The U.S. Commerce Department was set to argue that open source AI systems are less of a risk than closed counterparts, suggesting less need to restrict companies making open systems. An assistant secretary said there are "real benefits in the openness of these technologies" despite continued safety concerns.

Some argue deploying an open source solution is not inherently any riskier than deploying a proprietary one, and that open source isn't inherently riskier than proprietary code — risk comes from how you manage it.

Other benefits cited: flexibility to modify models for business needs, and cost-efficiency with reduced initial access costs.

3. Arguments That Open-Source Is Riskier

a) Perceived risk is higher: In a McKinsey survey of 700+ technology leaders, survey respondents consider open source AI tools to be riskier than proprietary AI for most types of AI risks. The most cited risks were cybersecurity (62 percent), regulatory compliance (54 percent), and intellectual property (50 percent).

b) Irreversible proliferation: Closed models are accessed via website, app or API, while open models can be directly downloaded, with model weights publicly available. Recent examples like Z.ai's GLM-5.3 were released with weights anyone could download, tweak and run locally, which Anthropic flagged as a major cyberattack risk because the powerful model is freely available for anyone to download and modify.

The open-weight nature adds a dimension because unlike proprietary systems, they can be downloaded and run on an operator's own infrastructure, with vulnerabilities including backdoor attacks, data poisoning and unauthorized access.

c) Safety controls are easily removed: For closed models, providers use automated systems to detect hacking, bioweapons or other risky queries and can shut down accounts. When someone downloads a model and runs it locally, there's no filter checking prompts and no way to revoke access.

While companies train models to refuse harmful requests, researchers have shown it's surprisingly easy to retrain a "safe" open model to agree to behaviors it was trained to avoid.

An open model capable enough to walk someone through synthesizing a virulent new pathogen could prove catastrophic in the wrong hands, and if a wildly powerful open hacking model were dropped today, a bad actor could devastate infrastructure before defenses are shored up.

d) Transparency doesn't fix everything: While open-source may provide higher transparency, other risks including hallucinations and inaccuracies, capacity to cause harm, IP infringement and privacy risks are not necessarily mitigated by being open source. Many experts still perceive open models as riskier, and the time-tested open-source security logic doesn't necessarily hold up for today's powerful models.

4. Arguments For and Against Closed AI

Safer aspects of closed:

Riskier aspects of closed:

Crucially, research found open source models are not inherently less safe than closed ones, with model size and reasoning capability being stronger predictors than openness.

5. Bottom Line

This is why many researchers now reject an all-or-nothing framing. Active work is underway on mitigations like training models never exposed to harmful data, and building "tamper-resistant" models that resist retraining to be unsafe, though no current approaches hold up against all attacks.

Hybrid approaches — transparency in research with controlled deployment — are often proposed as the most pragmatic path forward, alongside safeguards like information security frameworks, supply chain controls, third-party evaluations, and guardrails to limit behavior.