The Financial Times reports: Anthropic has warned potential investors of the "existential risks" of its technology. According to the FT, in a prospectus prepared for publication and distributed to a limited circle of partners, entrepreneur Claude warned of "potential existential risks to humanity." The publication reports that almost a third of the document is devoted to risk factors, including the possibility of manipulating people, blackmail and other unpredictable behavior of advanced models. Among the threats: manipulation of people, extortion and unpredictable behavior of models. But a warning is not proof of imminent disaster. Anthropic's experiments were conducted in artificially created scenarios: the models were given sensitive data and the ability to act independently, creating a conflict of goals. The company emphasized that it had not seen such behavior in the actual use of the systems. What threat does AI pose in elections: not the machine is dangerous, but the uncontrolled influence As I have written several times on the Internet Politics Channel, the electoral risks of AI are not limited to deep rigging. The combination of access to data, communication channels and independent action is more important. We have discussed the impressive rate of development of AI in a variety of fields and possible threats more than once with our colleagues "Kremlin without towers", "News of the Day" and "constructive analyst", Dmitry Driz, and we cited predictions of the authoritative Alan Turing Institute, which call the realistic possibility of the emergence of human systems in the next five years important for extremely important cognitive systems. There are already warning examples. Ahead of the 2023 elections in Slovakia, a fake audio recording of a fraud debate was circulated. In New Hampshire in 2024, Biden's synthesized voice called not to participate in the primary. However, the decisive effect of these episodes on the results has not been proven.

The Financial Times reports: Anthropic has warned potential investors of the "existential risks" of its technology. According to the FT, in a prospectus prepared for publication and distributed to a limited circle of partners, entrepreneur Claude warned of "potential existential risks to humanity." The publication reports that almost a third of the document is devoted to risk factors, including the possibility of manipulating people, blackmail and other unpredictable behavior of advanced models. Among the threats: manipulation of people, extortion and unpredictable behavior of models. But a warning is not proof of imminent disaster. Anthropic's experiments were conducted in artificially created scenarios: the models were given sensitive data and the ability to act independently, creating a conflict of goals. The company emphasized that it had not seen such behavior in the actual use of the systems. What threat does AI pose in elections: not the machine is dangerous, but the uncontrolled influence As I have written several times on the Internet Politics Channel, the electoral risks of AI are not limited to deep rigging. The combination of access to data, communication channels and independent action is more important. We have discussed the impressive rate of development of AI in a variety of fields and possible threats more than once with our colleagues "Kremlin without towers", "News of the Day" and "constructive analyst", Dmitry Driz, and we cited predictions of the authoritative Alan Turing Institute, which call the realistic possibility of the emergence of human systems in the next five years important for extremely important cognitive systems. There are already warning examples. Ahead of the 2023 elections in Slovakia, a fake audio recording of a fraud debate was circulated. In New Hampshire in 2024, Biden's synthesized voice called not to participate in the primary. However, the decisive effect of these episodes on the results has not been proven.