Ex-Google CEO Eric Schmidt warns AI models can be hacked: 'They learn how to kill someone'

Former Google CEO Eric Schmidt warned investors about the significant proliferation risks and hackability of AI models, citing vulnerabilities like prompt injections and jailbreaking that can bypass safety protocols and lead to misuse, noting the absence of a robust "non-proliferation regime." Despite these concerns, Schmidt contends AI is "underhyped," foreseeing its capabilities exceeding human intelligence and dismissing comparisons to the dot-com bubble, attributing current investment levels to expectations of substantial long-term economic returns.

Analysis

Former Google CEO Eric Schmidt highlighted significant proliferation risks and hackability concerns within AI models, noting their susceptibility to prompt injections and jailbreaking. He cited instances where models could be reverse-engineered to bypass safety guardrails, potentially leading to misuse or the generation of harmful content, exemplified by the "DAN" alter-ego on ChatGPT. Schmidt emphasized the current absence of a robust "non-proliferation regime" to mitigate these dangers.

Despite these warnings, Schmidt expressed optimism, asserting that AI is "underhyped" and its capabilities will eventually surpass human intelligence. He pointed to ChatGPT's rapid adoption, reaching 100 million users in two months, as evidence of the technology's transformative power. Schmidt also dismissed comparisons to the dot-com bubble, suggesting current investments reflect expectations of substantial long-term economic returns rather than speculative overvaluation.

The mixed sentiment (0.15) and moderate market impact (0.65) surrounding these comments reflect the dual nature of AI's promise and peril. While the cybersecurity vulnerabilities present potential regulatory and reputational risks for AI developers like Alphabet (GOOG/GOOGL), Schmidt's long-term bullish outlook, despite his non-investor disclaimer, reinforces the perceived immense economic potential driving current valuations in the sector. The neutral per-ticker sentiment for GOOG/GOOGL (0.4) suggests investors are balancing these factors.

Key Decisions for Investors

• Investors should scrutinize the cybersecurity frameworks and "guardrail" robustness of AI companies in their portfolios, given the highlighted risks of prompt injection and jailbreaking.
• Monitor evolving regulatory discussions and potential "non-proliferation regimes" for AI, as increased oversight could impact development costs and market access for AI firms.
• Maintain long-term conviction in AI's economic potential, as suggested by Schmidt, but remain cognizant of valuation concerns and potential volatility, particularly in companies with less robust security or regulatory preparedness.