Is open-source AI safer or riskier than closed/proprietary AI?
Is open-source AI safer — or riskier — than closed AI? I'll pull the latest evidence on both sides for you.
Neither is automatically safer - security and risk experts say Neither approach is automatically safer. The answer depends on what type of risk you mean: misuse by bad actors vs. transparency, concentration of power, and auditability.
Here is how the trade-off breaks down:
1. Why Open-Source / Open-Weight AI Is Seen As SAFER
a) Transparency and auditability: Open models publish their weights - the long series of numbers that encode internal structure - and sometimes code and training data. This allows outside researchers to inspect and verify what they are doing.
> The transparency of open-source frameworks often results in stronger security, as vulnerabilities are quickly identified and resolved by a global community of experts
> The belief that hiding source code makes a system safer is security through obscurity, and it usually fails
Signatories of a 2025 industry letter including Nvidia and Microsoft argued open weights broaden defensive capability, increase transparency, and allow vulnerabilities to be discovered and remediated across many teams.
By analogy to Linux, communities can more easily test open models for security flaws.
b) Privacy and local control: Because open models can be downloaded and modified, they offer advantages in cost, privacy, research transparency, and broader cyber-defense.
Businesses can run a local model on their own hardware, which may be the best way to keep their operations secure when processing sensitive data, and allow deep auditability for regulatory compliance and prevent vendor lock-in.
c) Decentralization of power: Stanford researchers note three advantages: distributing power, catalyzing innovation, and ensuring transparency
> When we use tools built on closed-source ChatGPT, we tacitly cede more control to OpenAI and Microsoft
> The clearest, biggest argument [for open models] is around preventing mass concentration of power—it is about making sure that this foundational technology is a public good
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 the closed models' security filters.
2. Why Open-Source Is Seen As RISKIER
a) No centralized control after release: Open weights enable outside testing and transparency, but they can also be modified and redistributed without the original developer's control.
Anthropic's head of international Guillaume Princen said: We need to be really careful with open-source on safety... by definition, there's more risk with them.
b) Easy removal of safeguards: Closed providers use automated systems to detect hacking/bioweapons queries and can shut down accounts. When someone downloads a model and runs it on their own hardware, there's no filter checking whether their prompts are safe, and there's no way to revoke their access.
Even if models are trained to refuse harmful requests, researchers have shown it's surprisingly easy to retrain a "safe" open model to agree to engage in the behaviors it was previously trained to avoid.
c) Attack surface: Because the code is publicly available, attackers can analyze it to identify vulnerabilities that they can exploit, such as poorly implemented security features or outdated dependencies.
A McKinsey/Mozilla/Patrick J. McGovern Foundation survey of 700+ technology leaders in 41 countries found Survey respondents consider open source AI tools to be riskier than proprietary AI for most types of AI risks. The most relevant AI risks cited include cybersecurity (62 percent of respondents), regulatory compliance (54 percent), and intellectual property (50 percent).
Stanford HAI researchers recognize certain areas where open-foundation AIs are indeed riskier than closed-foundation models, such as disinformation, bioweapons, hacking, and various forms of illicit pornography.
MIT Technology Review summarizes: Nevertheless, 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.
3. Why Closed / Proprietary AI Is Seen As SAFER
closed models typically demonstrate fewer safety and security risks, compared to open source, though they still demonstrate weaknesses and, crucially, are also vulnerable to jailbreaking techniques.thanks to their closed development process and exclusive access, and as of publication the top eight models on the Artificial Analysis LLM leaderboard are all closed.4. Why Closed AI Is Seen As RISKIER
The proprietary nature of closed-source models can obscure model transparency and make explainability difficult -- complicating AI governance and regulatory obligations for users.Closed platforms create a "black box" effect, hiding how decisions are made and limiting public oversight.
> You’re basically dealing with a black box, forced to trust the vendor’s word on how it works and what data it’s trained on
> Closed-foundation AI models may be safer in some ways, but they are also inherently more opaque... That is, in fact, its own form of risk
Closed-source models carry a high risk of vendor lock-in because users must accept the provider's exclusive decisions and Dependency on a single vendor can stifle flexibility and may lead to increased costs. Relying on proprietary services poses risks to long-term flexibility and data sovereignty.5. The Emerging Middle Ground
Experts increasingly reject an all-or-nothing framing. Current research focuses on making open models safer:
adversarial training, differential privacy (DP), and model sanitization techniques can effectively mitigate some of these risks, though achieving a balance between transparency and security remains a challengetamper-resistant models that resist retraining to remove safeguards, though no current approaches to making models tamper-resistant hold up against all tampering attacks.At the same time, many leaders expect hybrid adoption: Nearly three-quarters of respondents (more than 70 percent) say they are open to either open source or proprietary technologies, selecting models based on tradeoffs between accuracy, latency, cost, interpretability and security at different points in a workflow.
Bottom line:
Most governance proposals now aim not to ban either, but to manage their distinct failure modes: strong guardrails and monitoring for closed models, and community auditing, local deployment, supply-chain controls, and tamper-resistance research for open models.
Citations (3)
- Open models have real benefits. Do they outweigh the risks?
Backs: “transparency, concentration of power, and auditability. Here is how the trade-off breaks down: 1. Why Open-Source / Open-Weight AI Is Seen As SAFER. a) Transparency and auditability: Open models publish their weights - the long series of numbers that encode internal structure - and sometimes code and training data.”
- What is “Open” AI?
Backs: “Businesses can run a local model on their own hardware, which may be the best way to keep their operations secure when processing sensitive data, and allow deep auditability for regulatory compliance and prevent vendor lock-in. c) Decentralization of power: Stanford researchers note three advantages: distributing…”
- Open source
Backs: “Training models that never learn harmful knowledge, or building tamper-resistant models that resist retraining to remove safeguards, though no current approaches to making models tamper-resistant hold up against all tampering attacks. At the same time, many leaders expect hybrid adoption: Nearly three-quarters of…”