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

Your question sits at the center of AI's biggest safety debate — I'll map the arguments that open versus closed models are safer or riskier.

Open-source AI is not categorically safer or riskier than closed/proprietary AI - evidence suggests each model has different safety advantages and risks, and experts increasingly argue the risks should be assessed as marginal risk rather than a binary.

Arguments That Open-Source Is Safer

1. Transparency, auditability and peer review:
Open-source principles are argued to foster accessibility, collaboration, responsibility and interoperability. Proponents say open-source approaches lower barriers to innovation, accelerate science, and improve transparency.

Specific benefits claimed include:

The Nature Computational Science article notes that errors in open-source software are detected and corrected with everyday use, and much faster than in closed-source software. Implemented responsibly, open source can facilitate wide review to reinforce defenses. Linux and OpenSSL are open source yet power critical infrastructure, demonstrating security is achievable.

2. Prevents concentration of power:
When AI technologies like large language models are developed exclusively by a few companies, concentration of power over technology is known to hamper future innovation, fair competition and scientific progress. Proprietary AI could also jeopardize inclusiveness and responsibility, as companies may arbitrarily decide which countries and languages to support.

Cornell researcher David Widder argues the closed-AI safety narrative obscures other motives: companies such as OpenAI making arguments for closed AI systems on the basis of 'safety' have historically rejected other 'burdensome' safety mechanisms like audits, suggesting their argument is more about using regulation to entrench their forerunner position, not safety.

3. Secrecy does not necessarily improve security:
A 2024 study published in Science including Princeton and Stanford researchers concluded that limiting public access to underlying structures could stifle innovation, increase secrecy, and concentrate power.

The researchers 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. This is partly because even closed models can be subverted and partly because information for malicious actors might already be available on the internet.

They propose evaluating marginal risk: To what extent do open foundation models increase risk relative to closed foundation models, or to preexisting technologies such as search engines.

> "In some ways, the closed AI is a bigger risk simply because of the massive distribution," said Mike Linksvayer, vice president of developer policy at GitHub.

Arguments That Open-Source Is Riskier

1. Irreversible proliferation and misuse:
Once model weights are publicly released they cannot be recalled. Risks cited include vulnerabilities in open-source AI being public knowledge and use for nefarious purposes such as development of weapons and AI-generated propaganda campaigns.

A Department of Homeland Security report found that proliferation of publicly available AI tools could lower the barrier to entry for malicious actors seeking information on the composition, development and delivery of chemical and biological weapons. Similarly, as AI technologies advance, the lower barriers to entry for all actors may create novel risks from malign actors' enhanced ability to conceptualize and conduct CBRN attacks.

2. Safety guardrails can be removed:
A Stanford HAI policy brief highlighted that fine-tuning large language models on even a small number of harmful data points can easily disrupt safety mechanisms. This applies to open-weight models where users have full ability to fine-tune.

3. Software supply-chain vulnerabilities:
Like any open source project, AI models can harbour vulnerabilities, outdated components, or even hidden backdoors that scale with adoption.

Survey data cited by McKinsey shows respondents consider open source AI tools to be riskier than proprietary AI for most types of AI risks, with the most relevant risks cited as cybersecurity (62 percent), regulatory compliance (54 percent), and intellectual property (50 percent).

Other enterprise concerns include that open-source models can expose the unsuspecting to new threats, including manipulation of outputs and harmful content by bad actors.

Arguments That Closed/Proprietary Is Safer

1. Centralized control and monitoring:
Closed models are proprietary systems controlled by companies like OpenAI, Google, or Anthropic. If an enterprise prefers to offload environmental overhead and utilize robust, built-in safety filters with minimal upfront workforce investment, closed-source proprietary models are highly advantageous.

Developers can monitor usage, patch vulnerabilities, enforce terms of service, and revoke access — which is impossible with downloadable open weights.

2. Data privacy and compliance controls:
Leaders must weigh their need for strict data privacy, IP protection, and complete explainability — which can strongly favour local execution via open models. Conversely, closed models offer different trade-offs, but centralized providers often offer enterprise compliance certifications that self-hosted open models lack.

Arguments That Closed/Proprietary Is Riskier

1. Opacity and lack of external scrutiny:
Proprietary AI systems are difficult for members of the public to appraise, and thus to identify and fix errors within them. They introduce opaque "black-box" architectures, external data privacy risks, and operational token costs.

The 'open versus closed' binary obscures much: the most maximally open AI systems can provide some level of transparency but do not themselves 'democratize' access to AI, or enable outside scrutiny, as both require resources concentrated in the hands of a few large companies.

2. Single point of failure and market concentration:
As noted, closed ecosystems concentrate power, which critics say can crush free society concerns and reduce diversity of applications.

Synthesis: Beyond the Binary

Most recent policy analyses, including NTIA's report on open foundation models and the Stanford/AEI brief, recommend moving beyond the binary of open vs. closed model access and investing in evaluating marginal risk rather than resorting to blanket restrictions on release.

Authors in Nature argue that benefits are likely to outweigh downsides, especially if a responsible open-source approach with clear barriers against misuse is pursued. Proposed mitigations include:

In short: Open-source increases auditability and resilience but also increases proliferation risk; closed-source increases centralized oversight but also increases opacity and concentration risk. Regulators and enterprises often treat them as complementary, choosing open models for on-premise control, customizability and inspection, and closed models for managed safety filters and support — with governance applied to both.