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

Your question touches a lively debate on AI risk — I'll map the safety arguments for open and closed models before drawing a conclusion.

It is not a simple safer vs riskier binary - open-source and closed/proprietary AI have different safety profiles, and experts disagree on which is net safer. The consensus is: neither is automatically safer, both contain vulnerabilities.

The debate centers on where safety comes from:

•Open camp: safety through transparency - more eyes make it safer

•Closed camp: safety through control - fewer copies make it safer

Arguments that open-source is safer

1.Transparency and auditability. Proponents argue openness democratizes AI capabilities, enables independent safety auditing, prevents monopolistic concentration of power, and accelerates research.

2.Community review. Meta, which released Llama 3.1 405B as the first frontier-level open-source model, argues open-source AI is safer due to transparency and community review. Zuckerberg's reasoning is:

•More eyes on the code mean faster identification and resolution of potential issues

•Diverse applications reduce risk of a single dominant AI system

3.Anti "security through obscurity". The open-source community argues "security through obscurity" is a fallacy, and that keeping models closed ignores systemic risks that only a global community can identify and fix. From this perspective, open-source is actually safer because it eliminates the single point of failure and prevents a small group of engineers from having unilateral control.

4.Visibility of bugs. "We wash our dirty linen in public" - Many eyes make bugs shallow. Closed systems operate like black boxes where users have limited insight into how models function or how quickly vulnerabilities are disclosed.

5.Prevents concentration vs funds safety: The trade-off is often framed as:

•Innovation vs misuse

•Transparency vs control: Open systems can be audited, while closed systems can be throttled and monitored by the owner

•Competition vs concentration: Open models can prevent a few firms from controlling core AI, while proprietary models can fund huge training runs and safety teams

A Princeton Engineering review found that available evidence does not show that open models are riskier than closed models or information already available through standard research techniques like online searching.

A large arXiv review paper concluded overall the benefits of open-source Gen AI outweigh its risks, and encouraged open sourcing of models, training and evaluation data.

Arguments that open-source is riskier

1.Perceived risk is higher. In a McKinsey / Mozilla / Patrick J. McGovern Foundation survey of 700+ tech leaders, survey respondents consider open source AI tools to be riskier than proprietary AI for most types of AI risks. The most relevant risks cited include cybersecurity (62 percent), regulatory compliance (54 percent), and intellectual property (50 percent).

2.Irreversibility. Powerful models that anyone can download and modify can be used for cyberattacks, fraud, impersonation, deepfakes and misinformation. Safety guardrails can be stripped out after release, and a published model cannot be recalled.

Critics, including researchers at Anthropic, argue that releasing powerful model weights makes it impossible to prevent misuse — once weights are public, they cannot be un-released, and fine-tuning can remove safety guardrails. Dario Amodei has described this as "the asymmetry of open source: the benefits are diffuse but the risks can be concentrated".

3.Different threat models:

•Open-source: models can be vulnerable to traditional software supply-chain exploits, unvetted package dependencies, and local server misconfigurations

•Closed-source: insulates enterprises from infrastructure-level vulnerabilities, but exposes workflows to vendor-side outages, API key leaks, and even prompt jailbreaks

"Open source models such as DeepSeek and Llama are built on complex ecosystems of external libraries and dependencies... they can introduce significant vulnerabilities, with over 70% of applications containing open source flaws that often go undetected".

4.More mature governance for closed models. Proponents of closed-source argue they offer better security and control over potentially harmful applications, mitigating the risks of misuse or malicious exploitation, with built-in safeguards against misuse like content moderation and managed updates.

Bottom line for practical safety:

Dimension	Open / Open-Weight	Closed / Proprietary
Transparency	High, fully inspectable, auditable for bias/vulnerabilities	Low, "black box"
Misuse control	Low - weights cannot be retracted, guardrails can be fine-tuned away	High - API can be monitored, rate-limited, revoked
Security patching	Community-driven, fast if active community, but dependent on deployer to update	Vendor-controlled, faster time to value (48% say proprietary faster) but opacity
Privacy / Data control	Allows deep fine-tuning on private data without data leaving secure environment	Risk of data exposure to vendor
Regulatory compliance	Perceived as riskier (54%), but actually more audit-friendly under EU AI Act requiring auditability	Perceived as safer, but black-box makes audits harder

Researchers suggest "open" doesn't have to mean "no rules." A responsible release can include clear licenses, strong documentation, evaluation results, and guidance for safe deployment.

> For near/mid-term models with capabilities similar to today, the major paper on the topic argues benefits significantly outweigh marginal direct safety risks, and advocates against restricting development of open-source generative AI while encouraging regulation of improper use rather than development itself.

In short: If your top risk is unaccountable concentration of power, hidden bias, and lack of auditability, open-source is seen as safer. If your top risk is malicious misuse, proliferation of harmful capabilities, and irreversible release, closed is seen as safer. Most enterprises end up using a mix - more than 70% say they are open to either open source or proprietary technologies across the stack.

2.[Open source](https://www.mckinsey.com.br/~/media/mckinsey/business functions/quantumblack/our insights/open source technology in the age of ai/open-source-technology-in-the-age-of-ai_final.pdf)

Citations (2)

1arxiv.orgAcademic
Risks and Opportunities of Open-Source Generative AI
Backs: “A Princeton Engineering review found that available evidence does not show that open models are riskier than closed models or information already available through standard research techniques like online searching. A large arXiv review paper concluded overall the benefits of open-source Gen AI outweigh its risks, and…”
2mckinsey.com.brCompanies
Open source
Backs: “McGovern Foundation survey of 700+ tech leaders, survey respondents consider open source AI tools to be riskier than proprietary AI for most types of AI risks. The most relevant risks cited include cybersecurity (62 percent), regulatory compliance (54 percent), and intellectual property (50 percent).”