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

Comparing open versus closed AI safety — I'll gather current evidence on risks and benefits.

Open-source AI is not categorically safer or riskier than closed/proprietary AI — it is safer on some dimensions and riskier on others. Most experts describe tradeoffs rather than a single winner:

> Both approaches have safety tradeoffs. Open source allows broader scrutiny and research but can't be recalled if dangerous capabilities are discovered. Closed development enables more control but limits external safety research and concentrates power.

> Neither label guarantees safety.

Why open-source is argued to be SAFER

1. Transparency and auditability:
Open-source models allow users to see how the model works, making them more trustworthy and easier to audit for biases or errors. It enables public scrutiny where in a closed system the "safety" of a model is determined by a private board; in an open system, the global research community can audit the model for biases, vulnerabilities, and hallucinations.

Proponents argue openness democratizes AI capabilities, enables independent safety auditing, prevents monopolistic concentration of power, and accelerates research by allowing thousands of researchers to study model behavior rather than a handful at closed labs.

Mark Zuckerberg wrote in 2024 that he expects open source AI to be safer than the alternatives, especially against unintentional harm, because open systems can be widely scrutinized.

2. Faster vulnerability patching — "security through obscurity is a fallacy":
Conversely, the open-source community argues that "security through obscurity" is a fallacy. They contend that by keeping models closed, the industry is ignoring systemic risks that only a global community of researchers can identify and fix.

Empirical data from the history of software development proves that open-source systems identify and patch vulnerabilities significantly faster than siloed, proprietary alternatives. When we decentralize the development of AI, we distribute the capacity for oversight and defense across a global community of experts, rather than leaving our collective safety in the hands of a few executives. A transparent system is a self-healing system; a closed one is a fragile, single point of failure.

From this perspective, open-source AI is actually safer because it eliminates the single point of failure and prevents a small group of engineers from having unilateral control over the world's most powerful cognitive tools.

3. Privacy and control:
With closed AI, your prompts are sent to and processed on someone else's servers. With open source AI running locally, your data never leaves your own machine. For sensitive industries open source deployment can be significantly more private than any cloud-based closed model.

Closed models are not inherently safer: their operators can keep flaws from public scrutiny, monitor how people use them, impose restrictions, raise prices, or withdraw access.

Why open-source is argued to be RISKIER

1. Irreversibility and guardrail removal:
Critics, including some 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."

They are dangerous differently. Open-weight models can have their safety training removed for under $200 and can never be recalled.

Open Source fosters innovation, accessibility, and inclusivity but risks misuse, while Closed Source prioritizes safety, reliability, and polished experiences but can limit transparency and accessibility.

2. Evidence of criminal misuse:
Hackers and other criminals can easily commandeer computers operating open-source large language models outside the guardrails and constraints of the major artificial-intelligence platforms, creating security risks and vulnerabilities. Hackers could target the computers running the LLMs and direct them to carry out spam operations, phishing content creation or disinformation campaigns, evading platform security protocols.

A joint SentinelOne and Censys study over 293 days found 7.5% of observable system prompts could potentially enable harmful activity, and identified hundreds of instances where guardrails were explicitly removed, including use cases for hacking, hate speech and harassment, violent or gore content, personal data theft, scams or fraud.

Risks including hallucinations and inaccuracies, capacity to cause harm, infringement of intellectual property, and privacy risks are not necessarily mitigated in any manner by the fact that AI is an open source AI.

3. Safety eval comparisons are nuanced:
Several studies demonstrate that closed models typically demonstrate fewer safety and security risks, compared to open source. However, closed models still demonstrate weaknesses and, crucially, are also vulnerable to jailbreaking techniques.

All proved to be highly vulnerable to basic jailbreak techniques, with the models actioning harmful responses in between 90% and 100% of cases when the researchers performed the same attack patterns five times in a row.

Open systems can let outside researchers inspect, test, modify, and study models in ways closed systems may not permit. Safety and openness are related debates, but they are not the same question. A model does not become safe simply by being open, nor does a closed model become safe simply by keeping its weights private.

Bottom line

That is why the emerging consensus among researchers and policy groups is a mixed approach: some models open for research and scrutiny, others kept closed based on their capabilities and risks, weighing control vs convenience rather than assuming one label equals safety.