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

Is open-source AI the safer bet — or the riskier one? I'll weigh both sides and pull the latest evidence on risks and safeguards.

Open-source vs closed AI is not a simple "safer / riskier" binary - thoughtful evidence and experts land on both sides, and the answer depends on threat model, capability level, and governance. The widely used framing is actually a spectrum:

> A model can be fully open source (weights, training data, and training recipe all published), open weight (the trained weights released, but the data and recipe held back), or closed and proprietary (available only as a hosted service)

Most models marketed as "open source" today, including Llama, DeepSeek, Qwen and Gemma, are open-weight only.

Arguments that open-source / open-weight can be safer

1. Transparency and auditability:
Open Source AI is defined by the Open Source Initiative in October 2024 as requiring access to parameters, complete code, and enough information about training data that a skilled person could build a substantially equivalent system. Proponents argue this is needed to prevent corporate monopolies and ensure transparency and data privacy instead of relying solely on closed proprietary "black-box" models.

Publicly available model weights, code, and training methodologies allow independent researchers to audit models for algorithmic bias, security flaws, and hallucinations. The 2025 Foundation Model Transparency Index found that developers of open models are, on average, more transparent than closed developers.

Advantages cited include:

2. Security through openness:
By analogy to open-source software, tools like Linux are often seen as safer than closed-source counterparts because a vast community probes and patches vulnerabilities; similarly researcher communities can test open models for flaws

The myth that proprietary code is inherently more secure falls apart on inspection - hidden blueprints aren't a security posture. Kerckhoffs's Law takes the opposite view: a well-designed system should remain secure even when everything about it except the secret key is public knowledge

Coalitions including NVIDIA, Microsoft and Meta have argued open-weight models are safer than closed ones because a broad community can inspect them, identify vulnerabilities, and build safeguards, and that relying solely on closed models is not inherently safe

Arguments that open-source / open-weight can be riskier

1. Proliferation and irreversibility:
The case against wide-open weights is proliferation. Once weights are downloadable, they can be copied and redistributed beyond the developer's knowledge or control, and safety fine-tuning can be stripped back out

Once weights are released, there is no patch mechanism, no usage monitoring, and no ability to revoke access. A closed model can be updated, restricted, or shut down in response to discovered harms. An open model cannot

Open models turn AI from a metered service into a copyable artifact. Capability spreads at the speed of a download, and a release cannot be recalled

2. Easier removal of safeguards:
Anthropic argues that freely available models like GLM-5.3 pose serious cyberattack risks because they lack centralized safeguards and can be retrained to bypass safety controls

Similarly, Anthropic published a report arguing that GLM-5.3 poses major cyberattack risk because it is freely available to download and modify

Anthropic screens queries to closed models for hacking, biological weapons and other risks, and can shut down accounts. A downloaded model has no such filter and no way to revoke access. Companies also train models to refuse harmful requests - but researchers have shown it is surprisingly easy to retrain a "safe" open model to comply with behavior it was trained to avoid

The UK-led International AI Safety Report has warned that future open-weight releases are likely to aid even moderately skilled actors in cyber offense, and concerns extend to biosecurity and CBRN proliferation.

Why closed is not automatically safe

What independent assessments conclude

The US National Telecommunications and Information Administration (NTIA) studied this in 2024 and framed the question as one of marginal risk - the risk unique to open-weight models relative to closed models and existing technology

> It found the evidence too immature to conclude that open models pose more marginal risk than benefit. It recommended monitoring rather than restriction

> The evidence is genuinely mixed. Open weights raise proliferation concerns... But openness also enables independent auditing, red-teaming, and interpretability research that closed APIs block... the 2024 NTIA report framed this as a question of marginal risk and found the evidence too immature to declare open models riskier on balance

Many experts still perceive open models as riskier than closed models, especially as capabilities approach frontier levels. SaferAI and others conclude that models approaching frontier capability deserve far more systematic and independent safety testing before unrestricted release.

Bottom line

Most practical guidance therefore recommends a mixed, risk-based approach: use closed hosted APIs when you want highest capability with least operational overhead, use open-weight when data must stay on your infrastructure, deep fine-tuning or vendor independence is required, and accept that serving, scaling and security patching become your responsibility, with routing different work to different tiers rather than committing to one.