OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say

A Google-co-authored study of internet users in the U.S., Australia and the U.K., published in July, found that non-consensual AI-generated explicit content is often shared in private messages or clouds, making detection and reporting less likely. Google researchers said intervention at the source—the tools used to create fake sexual images—is critical, while recent red-teaming suggests some leading image generators from Google and OpenAI may be getting worse at blocking explicit outputs as their capabilities improve.

Analysis

The near-term exposure for Alphabet is more likely to be reputational and regulatory optionality than a material change to consolidated earnings: the key transmission is whether scrutiny shifts from individual abuse toward platform-level duties to prevent generation. That could raise compliance, monitoring and product-friction costs for Google’s image tools, while making safety performance a factor in enterprise and consumer trust. OpenAI and other image-generation providers face the same mechanism; better-verified safeguards could become a competitive differentiator, although compliance fixed costs may ultimately favor scaled providers over smaller rivals.

Time horizon: over days, headline risk is likely modest absent a named incident or regulator action. Over 1–3 months, watch for formal investigations, proposed rules, or independent testing that documents repeatable failures; these could widen the regulatory-risk discount across AI platforms. Over 6–18 months, the structural question is whether source-level controls become a durable operating requirement and constrain product capability or adoption. The contrarian case is that current concern may overstate direct financial exposure, while effective controls could protect trust and entrench incumbents. Conversely, company-led safety claims alone do not establish effectiveness. Reassess if regulators announce concrete enforcement or if independent evaluations show persistent failures despite product changes; a lack of such developments would weaken the bearish catalyst case.

Key Decisions for Investors

• No standalone short in GOOG on this report alone: the evidence describes a category-wide risk, without a quantified earnings channel or a company-specific regulatory action.
• Treat GOOG as a regulatory-risk watch item over the next 1–3 months. Escalate only on a formal inquiry, credible repeatable testing tied to Google products, or a material change in management disclosure on safeguards or compliance costs.
• If building AI exposure, favor providers that demonstrate independently verifiable controls rather than relying on company assertions; avoid assuming competitors are insulated, including OpenAI and other image-generation platforms.
• Falsify the risk thesis if independent testing shows durable improvement and no concrete regulatory steps emerge over the next few months; strengthen it if enforcement targets generation controls or Google identifies meaningful product restrictions or related expense.