Imani Vice President, Bright Simons has warned that growing public misunderstanding about how Artificial Intelligence (AI) systems improve their capabilities could lead to flawed policymaking and dangerous assumptions about the technology’s autonomy.
In a post published after the recent unveiling of Anthropic’s “Mythos” AI system, Simons who is also a technologist argued that much of the public reaction exaggerated the model’s intelligence while overlooking the extent to which advanced AI systems increasingly derive power from access to large networks of human-built digital tools.
“I fear that low awareness about HOW AI tools like ChatGPT get smarter will lead to poor societal thinking and decisions down the line,” Simons wrote.
The debate intensified after Anthropic said it was limiting the release of Mythos because of concerns about the system’s ability to identify vulnerabilities across software ecosystems. The company said the model scanned more than 1,000 open-source projects and identified more than 6,200 high- or critical-severity vulnerabilities among over 23,000 findings.
According to Simons, the results fueled fears that AI systems were rapidly approaching superhuman intelligence capable of replacing human decision-making across industries.
But he argued that the technology’s apparent leap in capability stemmed less from dramatic improvements in reasoning and more from expanded autonomy and persistent interaction with external tools, operating systems and software environments.
“AI has suddenly grown super-smart” is the wrong interpretation, Simons said, arguing that benchmark improvements over rival frontier models appeared relatively modest in several areas tied to human-style cognition, including search, knowledge retrieval and browser operation.
Instead, he said Mythos demonstrated strength in “persistent/long-horizon, wide-range, NON-AI TOOL use,” enabling it to autonomously combine and manipulate large numbers of external systems in ways difficult for humans to predict.
“That is right. We are giving AI tools autonomous control of a vast range of human-created tools,” he wrote.
The comments reflect growing global concern among regulators, technology executives and cybersecurity experts about so-called agentic AI systems capable of independently executing tasks across multiple digital platforms with minimal human intervention.
Simons argued that much of the world’s cybersecurity infrastructure was designed around the assumption that attacks would originate from humans rather than autonomous software agents capable of operating continuously at machine scale.
“Almost all of our software security was built in the belief that it is a human that will be trying to break in,” he said.
To illustrate the risks, Simons compared advanced AI systems connected to powerful digital infrastructure to “a highly energetic child” given access to missile systems, satellites and drones through a gaming console.
The growing sophistication of AI agents has intensified debate over how governments should regulate artificial intelligence while balancing innovation, cybersecurity and economic competitiveness.
Technology companies including OpenAI, Anthropic and Google are increasingly developing AI systems capable of carrying out complex multi-step tasks across software environments, raising concerns over control, accountability and systemic vulnerabilities.
Simons noted that policymakers and business leaders need a deeper understanding of how AI systems “harness human intelligence” rather than viewing them as independently omniscient technologies.
He said he plans to promote what he calls the “Social Edge Framework,” an initiative aimed at examining the policy and business implications of increasingly autonomous AI systems.
