The AI safety research community is increasingly questioning whether major AI companies like OpenAI and Anthropic are genuinely advancing safety or simply engaging in public relations. A new collaborative paper on "positive alignment" from Oxford, OpenAI, Anthropic, and Google has drawn sharp criticism from respected AI researchers, while geopolitical leaders are simultaneously pushing for AI guardrails.

What's Behind the Criticism of AI Safety Research?

Stephen Casper, an AI researcher, offered a blunt assessment of the recent "positive alignment" paper, stating that the work fails to meet academic standards. Casper noted the scholarship issues and questioned the research ethics involved, suggesting the paper functions more as corporate messaging than rigorous scientific inquiry. This critique reflects a broader tension in the AI field: as companies race to deploy increasingly powerful models, their safety research claims are facing heightened scrutiny from independent experts.

The criticism extends beyond a single paper. Researchers like Ryan Greenblatt are advocating for different approaches to AI safety, emphasizing the need for better model organisms and misalignment analogies rather than the current direction of industry-led safety initiatives. This suggests a fundamental disagreement about how to actually solve AI safety problems versus how to appear to be solving them.

How Are Geopolitical Leaders Responding to AI Risk?

In a striking development, both U.S. President Donald Trump and Chinese leader Xi Jinping are exploring AI guardrails, indicating that AI safety concerns have moved beyond the tech industry into high-level geopolitical discussions. This shift suggests that governments are beginning to view AI governance as a matter of national security and international relations, not merely a corporate responsibility issue.

The convergence of geopolitical interest and industry skepticism creates an unusual moment. While OpenAI and Anthropic present themselves as safety-conscious organizations, the actual effectiveness of their safety measures remains contested. At the same time, world leaders are recognizing that AI development requires some form of international coordination and guardrails.

Key Points in the AI Safety Debate

• Academic Credibility: Researchers are questioning whether major AI companies' safety papers meet legitimate academic standards or serve primarily as public relations materials.
• Research Direction: Independent AI safety researchers are advocating for different methodologies, such as studying model organisms and misalignment patterns, rather than relying on industry-led safety initiatives.
• Geopolitical Engagement: Both Trump and Xi Jinping are exploring AI guardrails, elevating AI safety from a corporate concern to a matter of international diplomacy and national security.
• Transparency Gap: There is growing concern that companies may be rehearsing safety claims without genuinely solving underlying alignment problems.

The broader context reveals a field in transition. OpenAI, which has positioned itself as a leader in responsible AI development, faces questions about whether its safety research translates into actual safeguards. Anthropic, often cited as the more safety-conscious alternative, is similarly under scrutiny. Meanwhile, researchers like Eliezer Yudkowsky have raised philosophical questions about whether large language models (LLMs), which are AI systems trained to predict and generate human language, might possess knowledge about human psychology that they cannot or will not reveal.

This tension between corporate claims and independent verification matters because the stakes are high. As AI models become more capable, the difference between genuine safety measures and performative ones could have significant consequences. The involvement of geopolitical leaders suggests that governments are beginning to recognize this gap and are taking steps to establish their own oversight mechanisms.

The coming months will likely reveal whether the AI industry's safety commitments are substantive or whether independent researchers and government bodies will need to establish alternative frameworks for ensuring AI development remains aligned with human values and interests.