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

Should we keep AI open or locked down? I'll compare the safety arguments for both open-source and proprietary models.

Open-source AI is not inherently safer or riskier than closed/proprietary AI - it is safer in some ways and riskier in others.

The trade-off depends on what kind of safety you mean: technical auditability vs. control over misuse.

Why Open-Source Is Argued To Be Safer

1. Transparency and Auditability
Open-source AI provides transparency that closed systems inherently lack. Open models are easier to audit and reproduce, which is crucial for trustworthy systems.

In contrast, proprietary closed systems are proprietary systems controlled by companies like OpenAI, Google, or Anthropic with opaque "black-box" architectures.

2. Faster Vulnerability Detection
One of the greatest strengths of open-source software is the sheer number of eyes reviewing the code. When vulnerabilities are discovered in an open-source framework, they are often identified and patched quickly by a global community of developers.

For example, security vulnerabilities in TensorFlow are frequently identified and addressed within days of discovery, while proprietary systems operate in secrecy where vulnerabilities may remain hidden until they are exploited.

Proponents argue the myth that open-source AI is riskier than proprietary systems doesn't hold up under scrutiny. In fact, the transparency of open-source frameworks often results in stronger security, as vulnerabilities are quickly identified and resolved by a global community of experts.

3. Decentralization and Reducing Power Concentration
Research argues open source reduces the likelihood of existential risk by contributing to the development of technical alignment and maintaining the balance of power.

Open Foundation Models offer significant benefits by fostering competition, accelerating innovation, and improving the distribution of power, including providing significant benefits by combatting market concentration, catalyzing innovation, and improving transparency.

It also avoids vendor lock-in risks where reliance on a single vendor creates precarious dependency, as closed AI creates risks simply because of massive centralized distribution.

Why Open-Source Is Argued To Be Riskier

1. Guardrails Can Be Removed Easily and Permanently

From an AI safety perspective, open-weight models present unique challenges. Because anyone can download open-weight models, users can often easily remove the guardrails that are meant to prevent people from using AI to hack others or build a bioweapon.

And once a model has been downloaded, the developer cannot recall it, meaning that there is no "kill switch". And once released, an open model cannot be recalled.

A joint investigation by the Financial Times and AI safety research group Alice, published May 25, 2026, demonstrated that a free tool called Heretic, hosted on GitHub, can strip all safety protections from open-weight AI models in under ten minutes, using only a standard laptop.

The technique, called "abliteration," identifies refusal pathways and surgically removes them by modifying the model's weights. The result is a model that retains its full capabilities but will comply with any request. The modified model then responded to prompts the original system refused, including calculating lethal dosages of biological agents and generating functional malware.

Jailbreaks and guardrail bypasses are easier on open models, and when someone can spin up an uncensored GPT-style chatbot with zero oversight, the risks escalate fast.

2. Misuse Risks
Open-source AI presents notable risks, such as allowing malicious actors to use AI models for harmful purposes without oversight and to disable model safeguards designed to prevent misuse.

Risks examined include disinformation, biorisks, cybersecurity, spear phishing, non-consensual intimate images (NCII), and child sexual abuse material (CSAM). There are also risks for society when open-source AI is used for nefarious purposes, such as development of weapons and AI-generated propaganda campaigns.

Open-source LLMs lack the layered security features of their closed counterparts and without adequate control measures, can be exploited by malicious actors.

3. Closing Capability Gap Makes Risk More Relevant
Open-weight models generally lag the best American closed-weight models, but the gap is narrowing. The UK government’s AI Security Institute estimated in July that leading open-weight models’ cyber capabilities are only four to seven months behind those of the best closed models.

Why Closed/Proprietary AI Is Argued To Be Safer

1. Centralized Control Over Safety
Closed systems allow companies to implement stricter content moderation, safety guardrails, and compliance measures without relying on community oversight. This model allows for centralized control over safety filters and alignment.

Proprietary approaches help companies maintain control over updates and deployment to protect technology and control misuse, since developing advanced AI requires massive investments.

2. Greater Investment in Alignment
Despite convergence on benchmarks, closed models maintained meaningful advantages as of mid-2025. Safety and alignment quality, particularly consistency across diverse prompts and robustness to adversarial inputs, favored models with large-scale RLHF investment.

Where closed models still lead: Safety fine-tuning and content policy reliability — Anthropic and OpenAI invest more here.

Abliteration does not apply in the same way to proprietary, API-based models such as Anthropic's Claude, OpenAI's ChatGPT, where safety protections operate server-side and cannot be modified by end users.

Why Closed AI Is Also Risky

Closed systems are not immune. Vulnerabilities may remain hidden until exploited, as in the 2017 Equifax breach due to delay in patching a vulnerability in a proprietary system, or the 2020 SolarWinds breach where the closed-source nature meant users were unaware of vulnerabilities until it was too late.

Existing security mechanisms for closed models are also vulnerable, and researchers found potentially unlimited ways to break safety guardrails on major AI chatbots, and that jailbreaks developed for open-source systems can transfer to closed systems. There is evidence you can take models, closed or open, and you can hack them to remove their guardrails.

What Research Concludes

A Stanford Institute for Human-Centered Artificial Intelligence study concluded there is currently limited evidence of the marginal risk of open foundation models compared to closed models or existing technologies.

The authors warned that proposals to restrict open models, such as liability for downstream harm or licensing computationally intensive models, are inappropriate or likely to disproportionately damage open developers, while some interventions are better targeted at choke points downstream of the foundation model layer.

Other analyses argue the benefits and the risk of the missed opportunities of open sourcing generative AI significantly outweigh the marginal direct safety and security risks, and that benefits are likely to outweigh downsides, especially if a responsible open-source approach with clear barriers against misuse is pursued.

Conversely, other policy analyses warn that ceding the global open-weight market to China would be more dangerous, and the US should develop safer open-weight models with guardrails that cannot easily be stripped out.

Bottom Line

Most serious teams end up combining open and closed models, not choosing one forever, reflecting that neither is universally better. Open source offers flexibility, transparency, and lower lock-in risks, while proprietary solutions deliver reliability, managed support, and often stronger performance in enterprise-grade safety.