Illinois SB 3444 AI 免責法案解析：OpenAI 力挺、90% 民眾反對，AI 造成大規模傷亡誰該負責？ | Illinois SB 3444 AI Liability Bill Explained: OpenAI Backs It, 90% of Public Opposes It
By Kit 小克 | AI Tool Observer | 2026-04-15
🇹🇼 Illinois SB 3444 AI 免責法案解析：OpenAI 力挺、90% 民眾反對，AI 造成大規模傷亡誰該負責？
Illinois SB 3444（又稱人工智慧安全法案）是目前美國最具爭議的 AI 立法之一。這項法案提議：只要 AI 公司有發布安全報告，就能免除因 AI 造成大規模傷亡或財產損失的法律責任。OpenAI 已公開支持，但民調顯示約 90% 的伊利諾伊州民眾反對。
Illinois SB 3444 是什麼法案？
SB 3444 是伊利諾伊州提出的 AI 安全法案，針對「前沿模型」（frontier model）設計。法案將「關鍵傷害」定義為：100 人以上死亡或重傷、10 億美元以上財產損失、或 AI 被用來開發化學、生物、放射性或核子武器。只要 AI 公司沒有故意或魯莽造成這些傷害，且有公開安全報告，就能獲得免責保護。
什麼是「前沿模型」？誰會被涵蓋？
法案定義：訓練成本超過 1 億美元算力的 AI 系統就算前沿模型。這個門檻會涵蓋所有主要 AI 公司，包括 OpenAI、Google、Anthropic、xAI 和 Meta。換句話說，這不是一般小公司的事，是整個 AI 產業頂層的遊戲規則。
OpenAI 為什麼支持 SB 3444？
OpenAI 已在州議會作證支持這項法案。AI 政策專家認為，這代表 OpenAI 的立法策略從「反對增加責任」轉向「主動爭取有條件免責」。對 AI 公司來說，這筆帳很清楚：發安全報告的成本遠低於面對天價賠償訴訟的風險。不過 WIRED 報導指出，SB 3444 比 OpenAI 過去支持的法案都更極端。
為什麼 90% 民眾反對？
根據辯論中引用的民調，約 90% 的伊利諾伊州民眾反對 AI 公司的免責條款。反對的核心理由很直覺：如果你的產品造成 100 人死亡，只要事先交了報告就不用負責？這聽起來像是讓大公司用合規文件換取免死金牌。批評者擔心這會降低 AI 公司認真對待安全的動機。
SB 3444 的合規替代方案
法案提供兩種合規替代路徑：（1）同意遵守歐盟的 AI 安全與安保要求；或（2）與美國聯邦機構簽署滿足特定要求的協議。此外，如果聯邦政府通過了涵蓋範圍重疊的法律，州法案將自動失效。這意味著 SB 3444 可能只是一個過渡方案。
這對 AI 產業有什麼影響？
如果 SB 3444 通過，它可能成為其他州效仿的模板，等於為整個美國 AI 產業設定了「有條件免責」的先例。這對新創公司和開源社群影響有限（因為訓練成本不到 1 億），但對頂尖 AI 實驗室來說，可能是一張極有價值的保險單。至於這張保險單夠不夠格，消費者和立法者顯然還有很大分歧。好不好用，試了才知道。
🇺🇸 Illinois SB 3444 AI Liability Bill Explained: OpenAI Backs It, 90% of Public Opposes It
Illinois SB 3444, also known as the Artificial Intelligence Safety Act, is one of the most controversial AI bills in the United States right now. It proposes that AI companies can be shielded from liability for catastrophic harms — including mass casualties and billion-dollar damages — as long as they publish safety reports. OpenAI has publicly testified in support, but polls show roughly 90% of Illinois residents oppose it.
What Is Illinois SB 3444?
SB 3444 is an AI safety bill targeting frontier models. It defines "critical harms" as: death or serious injury of 100+ people, $1 billion+ in property damage, or AI being used to develop chemical, biological, radiological, or nuclear weapons. AI companies that did not intentionally or recklessly cause such harms and have published safety reports would receive liability protection.
What Counts as a Frontier Model?
The bill defines any AI system trained with over $100 million in compute costs as a frontier model. This threshold covers all major AI labs: OpenAI, Google, Anthropic, xAI, and Meta. Smaller companies and open-source projects typically fall below this bar.
Why Does OpenAI Support SB 3444?
OpenAI testified before the Illinois legislature in favor of the bill. Policy experts see this as a strategic shift from opposing liability increases to actively seeking conditional immunity. WIRED notes that SB 3444 is more extreme than previous bills OpenAI has supported. The calculus is straightforward: publishing safety reports costs far less than facing catastrophic liability lawsuits.
Why Do 90% of Residents Oppose It?
A poll cited during legislative debates showed about 90% of Illinois respondents oppose AI liability exemptions. The core objection is intuitive: if your product kills 100+ people, should filing a safety report beforehand exempt you from accountability? Critics argue this creates a "compliance checkbox" that reduces AI companies' incentive to take safety seriously.
What Are the Compliance Alternatives?
The bill offers two alternative compliance paths: (1) agreeing to follow EU AI safety and security requirements, or (2) entering an agreement with a U.S. federal agency meeting specified criteria. If the federal government passes overlapping legislation, SB 3444 would automatically sunset. This positions the bill as potentially transitional.
What Does This Mean for the AI Industry?
If passed, SB 3444 could become a template for other states, effectively establishing a "conditional immunity" precedent for frontier AI in the U.S. The $100M compute threshold means startups and open-source projects are unaffected, but for top AI labs, it could be an extremely valuable insurance policy.
Sources / 資料來源
- Quartz: OpenAI backs Illinois bill to shield AI firms from harm lawsuits
- DNYUZ: OpenAI Backs Bill That Would Limit Liability for AI-Enabled Mass Deaths
- LegiScan: Illinois SB3444 Official Bill Text
常見問題 FAQ
Illinois SB 3444 是什麼？
SB 3444 是伊利諾伊州提出的 AI 安全法案，允許前沿 AI 模型開發商在公開安全報告的前提下，免除因 AI 造成大規模傷亡或財產損失的法律責任。
SB 3444 的前沿模型門檻是什麼？
訓練成本超過 1 億美元算力的 AI 系統就算前沿模型，涵蓋 OpenAI、Google、Anthropic、xAI、Meta 等主要 AI 公司。
SB 3444 的關鍵傷害定義是什麼？
100 人以上死亡或重傷、10 億美元以上財產損失、或 AI 被用來開發化學/生物/放射性/核子武器。
OpenAI 為什麼支持 SB 3444？
OpenAI 認為有條件免責（發布安全報告即可）能平衡創新與安全。政策專家認為這是 OpenAI 立法策略的重大轉向。
SB 3444 會影響開源 AI 嗎？
不太會。1 億美元的訓練成本門檻意味著大部分開源專案和小型公司不在法案涵蓋範圍內。
延伸閱讀 / Related Articles
- Microsoft GigaTIME 開源 AI 癌症診斷：10 美元病理切片變 2000 美元免疫螢光掃描，4000 萬細胞訓練、24 種癌症通吃 | Microsoft GigaTIME Open-Source AI Cancer Tool: Turns $10 Slides Into $2000 Immune Scans Across 24 Cancer Types
- ChatGPT 廣告 6 週賺破 1 億美元：OpenAI 預估 2026 廣告營收 25 億，2030 衝 1000 億 | ChatGPT Ads Hit $100M in 6 Weeks: OpenAI Projects $2.5B Ad Revenue in 2026, $100B by 2030
- Generalist AI GEN-1 機器人基礎模型：99% 成功率、3 倍速度，只需 1 小時數據就能學會新任務 | Generalist AI GEN-1: 99% Success Rate, 3x Faster, Masters New Tasks With Just 1 Hour of Data
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言