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

This is a genuinely contested question in AI policy circles, with serious arguments on both sides rather than a settled answer. Let me pull in some current perspectives to give you a well-rounded view.# Open-Source vs. Closed AI: There's No Clean Answer—It Depends on What Risk You're Weighing

This is one of the most actively contested debates in AI policy, and reasonable experts land on opposite sides using the same evidence. Here's the honest breakdown.

The Case That Open (Weight) Models Are Riskier

Irreversibility and loss of control. Once weights are released, the model can be run locally and no one can rescind access once the weights are out, or monitor how anyone uses them. Closed models hide their architecture but rely on a few certified experts and internal audits, while open ones expose everything—including to bad actors.

Guardrails can be stripped. Safety training is optional once weights are public. Researchers at Palisade Research showed they could undo the safety tuning on Meta's Llama 2-Chat for under $200 on a single GPU, taking the model from refusing almost all harmful requests to refusing about 1% of them. This "abliteration" is now widespread—every open-weight model has an abliterated version, and Hugging Face lists more than 8,000 models under the search term "abliterated".

Anthropic's position. Anthropic argues open-weights models do potentially present a higher risk than closed models, because it is very difficult to apply guardrails to them or monitor their usage, and once weights are released they cannot be withdrawn. They're particularly worried about biosecurity: sufficiently capable models may be able to quickly weaponize pandemic-level viruses with widely available materials, whereas defense against these agents is a multi-year operational task in the best case.

Oversight gap. Unlike closed-weight models, open-weight models can be run on a user's own hardware rather than on a company's cloud, making insight into how they're being used more challenging.

The Case That Open (Weight) Models Are Safer or Comparable

The "many eyes" argument. Some experts argue, by analogy to open-source software, that open models have safety advantages over closed models—open-source tools like Linux are often seen as safer because their vulnerabilities have been probed and patched by a vast community of online contributors.

Closed models aren't actually safe by default. It would be a vast oversimplification to deem open models dangerous and closed models safe—the defenses AI companies build around their closed models are far from perfect, and people misuse those models all the time. Microsoft makes a similar point: relying solely on closed models is not inherently safe—they can be breached, misused, or fail in ways outsiders can't detect, and concentrating advanced AI capabilities behind a few closed models compounds that risk.

Industry voices are shifting. Notably, more American AI leaders are describing Chinese open-weight models as better for safety and security than closed-source models—AI pioneer Andrew Ng said open-weight models "seem safer to me than closed-weight models", challenging the standard narrative. Over 70 companies including Google, Microsoft, and Nvidia wrote that open models "broaden defensive capability, increase transparency, and allow vulnerabilities to be discovered and remediated."

Concentration-of-power risk. A different kind of "safety" concern: "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," says Elizabeth Seger of the Tony Blair Institute, noting "we're at this juncture where the question of who controls the infrastructure that allows AI to operate is as big of a question as who gets to control information on the internet."

Where the Real Consensus Is Forming

Most serious analysts now reject the binary framing entirely:

Bottom Line

Neither category is "safer" in an absolute sense—each shifts risk to different places:

The most credible emerging view (e.g., from Anthropic, Brookings, and policy researchers) is that safety should be evaluated per-model through actual testing and red-teaming, not assumed from the open/closed label alone — and that a healthy AI ecosystem probably needs both.