We are experiencing a rapid shift in artificial intelligence technology. AI systems are advancing faster than ever, and many organizations claim they are making these systems safer. But is that really true? Or are we being fooled into thinking everything is under control when it might not be? Just like airport security measures that look strict but don’t stop real threats, AI safety efforts might be more about appearances than actual safety.

The Illusion of Progress in AI Safety

Companies develop techniques that seem to improve AI models. These models refuse harmful requests more often and score better on helpfulness tests. They publish papers and release press statements that show progress. This creates a sense that AI is becoming safer and more reliable. But these visible improvements might not reflect the real risks posed by increasingly powerful AI systems.

The truth is, many of these safety improvements focus on test results rather than addressing fundamental safety concerns. The measures often create a false sense of security. They give the impression that AI is safer than it really is, which can lead to more deployment without proper safeguards. This pattern is similar to security theater—actions that look effective but don’t actually prevent threats.

The Rapid Pace of AI Development and the Funding Gap

In 2025, AI models are being released at an unprecedented rate. For example, GPT-4o came out in May 2024, and GPT-5 followed in August 2025. Similarly, Anthropic released Claude 3.5 in June 2024 and Claude 4 in May 2025. The time between major releases has shrunk by about 43%, and each new model has capabilities that introduce new risks.

At the same time, safety research is severely underfunded. By mid-2025, only around $67 million had been allocated to AI safety efforts this year. Yet, venture capitalists poured nearly $193 billion into AI startups in 2025 alone. This huge influx of investment is mainly focused on building new AI systems, not on making them safer. The global AI safety research ecosystem receives only about $600-650 million annually—just a tiny fraction of overall AI funding.

This funding gap is dangerous. Without enough resources dedicated to safety, we risk deploying highly capable AI systems that might have serious unforeseen risks. The current focus on visible progress in AI safety could be distracting us from addressing the real dangers. We need to evaluate whether the metrics we use truly measure safety or just give us a false sense of security.

Investing more in AI safety research and developing transparent, safer AI systems should be a priority. The future of AI depends on whether we can align its development with real safety measures, not just appearances. Otherwise, we risk facing consequences we are unprepared for.

Inspired by

• https://justainews.com/ai-compliance/ai-ethics-and-society/why-ai-safety-research-making-things-worse/