**Anthropic Identifies Biological Weapon Risks Linked to AI Research**
In a recent threat intelligence report, Anthropic, an AI research company, disclosed alarming instances of its AI models being used in research that could potentially aid in the development of biological weapons. The report, published on Thursday, highlighted five specific cases involving researchers who utilized the AI model Claude for projects that raised significant security concerns.
Among the troubling examples, one case involved a military research institute engaged in gain-of-function research aimed at enhancing the chikungunya virus's transmissibility and its ability to evade the immune system. This type of research has been controversial due to the potential risks associated with creating more virulent pathogens.
Another instance cited in the report involved a scientist researching highly pathogenic avian influenza (bird flu) and exploring ways for the virus to adapt to mammalian hosts. Additional cases included work on viruses related to smallpox and efforts to redesign hazardous toxins.
Anthropic noted that some researchers attempted to circumvent the company's restrictions by concealing their identities, routing their internet traffic through other countries, and utilizing services designed to bypass safety controls. In response to these activities, Anthropic has banned the accounts involved and shut down the services that facilitated these breaches of security.
While the company refrained from definitively stating that these scientists were attempting to create biological weapons, it expressed concern that sophisticated actors might disguise malicious intentions behind research that appears legitimate. Anthropic acknowledged the increasing capabilities of its AI systems, stating that earlier models were insufficient for aiding advanced users in dangerous biological research, a claim that no longer holds true for its current models.
The report's release follows a notable experiment in which U.S. scientists successfully used AI to design complete viral genomes, resulting in the creation of 16 functioning viruses that had not previously existed in nature. Although these viruses infected bacteria rather than humans, researchers have cautioned that this experiment underscores the rapid advancements in AI's biological capabilities.
In addition to the biological weapon risks, Anthropic's report outlined other alleged misuses of its AI model, including applications in cyberattacks, surveillance, propaganda, fraud, and support for conventional weaponry such as missiles and drones.
The timing of the report coincides with recent comments made by Anthropic researcher Evan Hubinger, who suggested that there is a greater than 10% chance that AI could pose an existential threat to humanity within the next decade. Hubinger's remarks came after the resignation of fellow researcher Jacob Coxon, who criticized both Anthropic and OpenAI for allegedly neglecting the potential dangers associated with rapidly advancing AI technologies. Coxon warned that sufficiently advanced AI could not only hack critical infrastructure but also assist in the creation of "extinction-level bioweapons."
As the conversation around AI's capabilities and ethical implications continues to evolve, Anthropic's findings serve as a stark reminder of the potential risks associated with the misuse of advanced technologies in sensitive areas such as biological research. The company emphasizes the need for vigilance and robust safeguards to prevent the exploitation of AI in harmful ways.