This is a genuinely contested question among AI researchers, policymakers, and industry leaders, with reasonable arguments on both sides. Rather than searching for a definitive "answer" (since this is fundamentally a matter of ongoing debate rather than settled fact), let me lay out the core arguments, since I'm confident about the shape of this debate from general knowledge.
Arguments that open-source AI is safer
•Transparency enables scrutiny. More eyes on the code/weights means more people can find vulnerabilities, biases, and failure modes than a closed team could alone.
•Reduces concentration of power. If only a few corporations control the most powerful AI, society becomes dependent on their judgment, incentives, and security practices. Open models distribute capability more broadly.
•Enables independent safety research. Academics and third-party auditors can't rigorously study a model's internals, robustness, or alignment properties if they can't access weights or training data.
•Avoids "security through obscurity." Closed models have sometimes been jailbroken or leaked anyway (e.g., early leaks of proprietary model weights), suggesting secrecy isn't a reliable safety mechanism.
•Competitive/geopolitical argument. Proponents argue that if democratic countries don't lead in open models, authoritarian states or less safety-conscious actors will fill the gap.
Arguments that open-source AI is riskier
•Irreversibility. Once weights are released, there's no way to patch, monitor, or revoke access if dangerous capabilities or misuse are discovered later—unlike an API-gated model, which can be restricted or shut down.
•Removes safety guardrails. Fine-tuning can strip out safety training relatively cheaply, so "safety alignment" in open models is easier to undo than in closed systems protected by API-level filtering.
•Misuse at scale. Bad actors (state or non-state) can use open models for things like bioweapons design assistance, disinformation, or cyberattacks without any centralized ability to detect or stop misuse.
•Proliferation to adversaries. Open models can be freely obtained by sanctioned states or groups, whereas closed providers can enforce export controls or usage policies.
•Harder to track harm. With closed APIs, providers can monitor usage patterns and intervene; with downloaded weights running locally, there's no telemetry or oversight at all.
Where the debate actually stands
Neither side has "won." It often splits along:
•Capability level: Many argue the calculus differs for current frontier models (moderate risk) versus hypothetical future systems with much more dangerous capabilities (e.g., bioweapon uplift), where irreversibility concerns dominate.
•Type of risk: Open-source may reduce concentration/misuse-by-monopoly risks while increasing diffuse misuse risks.
•Who you trust more: Those skeptical of corporate/government control of AI tend to favor open models; those more worried about catastrophic misuse by bad actors tend to favor closed, controllable systems.
Would you like me to dig into current specifics—like recent policy positions (e.g., U.S. or EU stances), notable incidents, or specific company arguments (Meta's Llama vs. OpenAI/Anthropic's closed approach)? I can search for the latest developments if that would help.