> ## Content Index
> Fetch the complete content index at: https://www.thedelatorrereview.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Anthropic’s Advanced AI Framework and Its Alignment With California and EU Model Rules
- URL: https://www.thedelatorrereview.com/anthropics-advanced-ai-framework-and-its-alignment-with-california-and-eu-model-rules/
- Published: 2026-10-24T20:40:50.000Z
- Updated: 2026-10-26T20:37:50.000Z
- Description: Anthropic proposes enforceable oversight of frontier AI. Its framework shares features with California’s TFAIA and the EU AI Act, while differing in coverage, risk definitions, transparency, and government powers.
- Author: L. De La Torre
- Tags: California, AI model, Foundation/Frontier Model, Floating-Point Operation (FLOP), AI Developer, Transparency, Audits, General-Purpose AI (GPAI), Unacceptable Risk / Prohibited AI Practices, High-Risk AI

In June 2026, Anthropic published its [**Advanced AI Framework**](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com), a proposal for government oversight of catastrophic risks from the most powerful AI models. Presented in [Policy on the AI Exponential](https://www.anthropic.com/policy-on-the-ai-exponential?ref=thedelatorrereview.com), the framework calls for frontier developers to test their models, disclose risks and safety practices, obtain independent evaluations, and maintain robust security. It also proposes government authority to block or deter dangerous deployments, paired with safeguards against regulatory overreach. 

The announcement pairs that safety agenda with an [Economic Policy Framework](https://www-cdn.anthropic.com/files/4zrzovbb/website/9ea607a5dd67c168093829b701f3a0a6d21156d5.pdf?ref=thedelatorrereview.com) addressing how to prepare workers and the economy for AI’s effects and distribute its financial benefits more broadly. Written primarily for the US federal government, **it argues that public disclosures alone are no longer sufficient as model capabilities advance.**

---

> **The Advanced AI Framework focuses on two complementary tasks: regulating frontier developers and strengthening society’s resilience to biological and cyber threats.**

---

Anthropic’s proposal follows two legislative developments. The [EU AI Act](https://ai-act-law.eu/?utm%5Fsource=gdpr-info.eu) received final Council approval on **May 21, 2024**, and entered into force on **August 1, 2024**, establishing obligations for providers of general-purpose AI models, with additional requirements for models presenting systemic risk. California subsequently enacted its Transparency in Frontier Artificial Intelligence Act (TFAIA) through [SB 53](https://leginfo.legislature.ca.gov/faces/billHistoryClient.xhtml?bill%5Fid=202520260SB53&ref=thedelatorrereview.com), signed on **September 29, 2025**. Anthropic published its [proposed framework](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com) in **June 2026**, adding recommendations for further government oversight. The three approaches overlap, but “frontier model,” “general-purpose AI model,” “catastrophic risk,” and “systemic risk” identify different categories, with distinct coverage, obligations, and enforcement mechanisms.

- California requires covered developers to formalize their safety practices and, particularly for the largest developers, stand behind those commitments. See, [California’s Frontier AI Transparency Act (TFAIA): Building an Ongoing Safety and Oversight System for Powerful General-Purpose AI Models](https://www.thedelatorrereview.com/californias-frontier-ai-transparency-act-building-an-ongoing-safety-and-oversight-system-for-powerful-general-purpose-ai-models/)
- The EU combines general model-provider obligations with additional duties for models presenting systemic risk. See [General-Purpose AI Models Under the EU AI Act: Classification, Obligations, and Systemic Risk](https://www.thedelatorrereview.com/general-purpose-ai-models-under-the-eu-ai-act-classification-obligations-and-systemic-risk/).
- Anthropic proposes a targeted catastrophic-risk regime that adds external evaluation and intervention powers to transparency obligations.

**This article examines Anthropic’s proposed Advanced AI Framework and compares it with California’s TFAIA and the EU AI Act’s rules for general-purpose AI models.** It explains where the three approaches align, where their coverage and obligations differ, and what those differences mean for model evaluation, transparency, risk management, and government oversight.

# What Anthropic Proposes

The framework combines obligations for frontier developers with investments in societal resilience. Its priority risks are biological weapons, offensive cyber operations, loss of control, and automated research and development that could amplify those dangers. (See, [Anthropic’s policy overview](https://www.anthropic.com/policy-on-the-ai-exponential?ref=thedelatorrereview.com))

Its proposed developer obligations include:

- Publishing and following a safety framework, with annual compliance certification. ([Framework, p. 5](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=5).)
- Publishing system cards for qualifying deployments and public risk reports at least every six months. ([Framework, p. 6](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=6).)
- Reporting critical safety incidents to a designated agency within 15 days of discovery or awareness of facts supporting a reasonable belief that an incident occurred. ([Framework, p. 7](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=7).)
- Engaging a qualified independent evaluator with access to the developer’s most capable models and unredacted risk reports and system cards. ([Framework, pp. 7–8](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=7).)

The framework also expressly references SB 53 when recommending a catastrophic-risk definition. ([Framework, pp. 4–8](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com) Footnote 2)

- Security would extend across the development environment. Government would have authority to block or deter dangerous deployment, subject to safeguards against overreach. Resilience investments would strengthen biological defenses and cybersecurity beyond the developers themselves. ([Anthropic’s policy overview](https://www.anthropic.com/policy-on-the-ai-exponential?ref=thedelatorrereview.com))

# Who Would Be Covered

![](https://storage.ghost.io/c/54/ef/54efeb65-4f2f-479e-b4bb-a9ee526feeaa/content/images/2026/10/Antique-map-style-AI-coverage-poster.png)

Compute thresholds are a useful starting point, but they serve different functions.

- **Anthropic’s proposal.** Coverage would require both a model trained using **more than 10²⁵ floating-point operations (FLOP)** and a developer earning **more than $500 million in annual AI-derived revenue** or spending **more than $1 billion annually on AI research and development**. ([Framework, p. 3](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=5).)
- **California’s TFAIA.** A foundation model becomes a frontier model when its training uses **more than 10²⁶ integer or floating-point operations**. Baseline obligations apply to all frontier developers. Enhanced obligations apply to large frontier developers—those that, together with their affiliates, exceeded **$500 million in annual gross revenues during the preceding calendar year**. (See [Cal. Bus. & Prof. Code § 22757.11](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=BPC§ionNum=22757.11.&ref=thedelatorrereview.com).)
- **EU GPAI rules.** The starting point is whether a model displays **significant generality and can competently perform a wide range of distinct tasks**. Compute alone does not define a general-purpose AI model. The applicable provider obligations arise from placing a covered GPAI model on the EU market, without an equivalent revenue threshold. (See, [Article 3(63) ](https://artificialintelligenceact.eu/article/3/?ref=thedelatorrereview.com)of the EU AI Act.)
- **EU systemic-risk tier.** Training compute exceeding **10²⁵ FLOP** creates a **rebuttable presumption of high-impact capabilities**. The European Commission can also classify a model as presenting systemic risk based on equivalent capabilities or impact, including where the compute threshold is not met. (See, [European Commission’s 2025 GPAI Guidelines](https://digital-strategy.ec.europa.eu/en/library/guidelines-scope-obligations-providers-general-purpose-ai-models-under-ai-act?ref=thedelatorrereview.com) and [Articles 51-52](https://ai-act-law.eu/chapter/5/?ref=thedelatorrereview.com) of the EU AI Act)

The shared 10²⁵ figure does not make Anthropic’s proposed scope equivalent to the EU systemic-risk category. The EU can capture models below that threshold, and a provider’s revenue does not determine whether its model presents systemic risk.

Likewise, California’s higher compute threshold does not mean its law is uniformly narrower. California imposes baseline obligations on frontier developers that do not satisfy its large-developer revenue threshold. Anthropic’s financial criteria govern entry into its proposed developer regime.

> **Practice Tip:** Determine coverage separately under each framework. Record the model’s capabilities, relevant training compute, provider role, market availability, and the applicable financial measure. A single conclusion that a model is “frontier AI” cannot replace those assessments.

# How the Proposal Aligns With CA TFAIA

## (1) Enforceable Safety Commitments

TFAIA’s central governance requirement applies to large frontier developers. They must write, implement, comply with, and publicly disclose a frontier AI framework.

- That framework must address risk thresholds, assessments, mitigations, deployment decisions, third-party assessment, cybersecurity for unreleased weights, incident response, internal governance, and risks from internal use. Developers must review it at least annually and publish material modifications with a justification within 30 days. (Cal. Bus. & Prof. Code [§ 22757.12(a)–(b).](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.12.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com))
- California therefore already goes beyond asking developers to describe their practices. Failure to comply with a developer’s own framework can trigger Attorney General enforcement. Large frontier developers may face penalties of up to $1 million per violation for specified failures. (Cal. Bus. & Prof. Code [§ 22757.15.](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.15.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com))

Anthropic’s proposed annual certification would add another accountability mechanism. ([Framework, p. 5](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com))

The practical overlap is substantial. 

---

> **A credible safety program requires documented criteria, evidence supporting risk assessments, governance over release decisions, and records showing that the organization followed its commitments.**

---

## (2) Public Disclosures and Internal Use

TFAIA requires transparency reports before or concurrently with deployment of a new frontier model or a substantially modified version. Large frontier developers must include summaries of catastrophic-risk assessments, results, third-party involvement, and other steps taken under their framework. These disclosures may form part of a system card or model card. (Cal. Bus. & Prof. Code [§ 22757.12(c)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.12.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com).)

California also addresses risk before public deployment. Large developers must transmit summaries of assessments of catastrophic risk arising from internal use to the Office of Emergency Services, generally every three months or on another reasonable schedule communicated to OES. (Cal. Bus. & Prof. Code [§ 22757.12(d)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.12.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com).)

---

> **This is a significant point of alignment with the broader concern about increasingly capable models operating inside AI laboratories. Internal use belongs in the safety analysis even when customers cannot access the model.**

---

## (3) Independent Evaluation and Deployment Restrictions

TFAIA requires a developer’s framework to explain its approach to using third parties. It also requires disclosure of the extent of their participation. Those requirements do not establish a universal obligation to obtain an independent audit. (Cal. Bus. & Prof. Code [§ 22757.12(a)(5), (c)(2)(C)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.12.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com).)

- Anthropic proposes a more explicit independent-evaluation requirement and authority to restrict dangerous deployment. ([Framework, p. 7-9](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com))

---

> **The distinction is between enforcing a developer’s governance commitments and creating an additional mechanism (the independent evaluation) for public authorities to act on unacceptable residual risk.** 

---

TFAIA’s penalty provision should not be viewed as a general model-approval or licensing system.

## (4) Similar Incident Reporting Does Not Mean Identical Triggers

California generally requires critical safety incidents to be reported to OES within 15 days of discovery. An incident presenting an imminent risk of death or serious physical injury instead requires disclosure within 24 hours to an appropriate authority. (Cal. Bus. & Prof. Code [§ 22757.13(c)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.13.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com).)

Anthropic similarly proposes a 15-day reporting period, with the clock starting upon discovery of either a critical safety incident or facts supporting a reasonable belief that one occurred. Reports would go to a designated government agency. Unlike California’s rule, the proposal does not specify a separate 24-hour deadline for imminent threats to life or serious physical injury. (Framework, [p. 7](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=7).)

---

> **The same headline reporting period can conceal important differences. A compliance process must identify the applicable incident definition, discovery trigger, recipient, urgent escalation rule, and required contents.** 

---

# How the Proposal Aligns With the EU AI Act

## (1) The Closest Comparison Is the Systemic-Risk Tier

Chapter V of the EU AI Act creates two layers of model-provider obligations.

**(1) Article 53 generally requires providers of covered GPAI models to maintain technical documentation, provide information to downstream system providers, adopt an EU copyright-compliance policy, publish a training-content summary, and cooperate with regulators.** Limited accommodations apply to qualifying open-source models without systemic risk.

**(2) Article 55 adds duties for providers of GPAI models with systemic risk:**

- Model evaluation, including documented adversarial testing.
- Assessment and mitigation of systemic risks.
- Tracking and documenting serious incidents, and reporting relevant information and possible corrective measures without undue delay.
- Adequate cybersecurity for the model and its physical infrastructure.

See, [General-Purpose AI Models Under the EU AI Act: Classification, Obligations, and Systemic Risk.](https://www.thedelatorrereview.com/general-purpose-ai-models-under-the-eu-ai-act-classification-obligations-and-systemic-risk/)

---

> **The closest functional comparison with Anthropic’s proposal is Article 55’s safety regime. Comparing the proposal only with the EU’s baseline GPAI obligations would miss that additional layer.**

---

## (2) Systemic Risk Extends Beyond Catastrophic Risk

**California’s catastrophic-risk definition is tied to specified conduct and a foreseeable, material risk** of more than 50 deaths or serious injuries, or more than $1 billion in property damage or loss, arising from a single incident. Statutory exclusions further limit the category. (Cal. Bus. & Prof. Code §§ [22757.11(c)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.11.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com), [22757.16.](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?sectionNum=22757.16.&nodeTreePath=10.62&lawCode=BPC&ref=thedelatorrereview.com))

**Anthropic likewise focuses on catastrophic risk**, defining it as a foreseeable and material risk that the development, storage, use, or deployment of a covered model will materially contribute to significant death, injury, or damage. Although a footnote recommends alignment with California’s SB 53 (i.e. TFAIA), the proposal does not expressly reproduce California’s numerical harm thresholds. It prioritizes biological weapons, offensive cyber operations, loss of control, and automated research and development that could accelerate or amplify those risks. (Framework, [p. 4, including footnote 2](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=4).)

EU systemic risk is broader. [Article 3(65)](https://ai-act-law.eu/article/3/?ref=thedelatorrereview.com) of the EU AI Act addresses risks arising from high-impact capabilities that can propagate across the value chain, including actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or society as a whole.

---

> **A risk can warrant EU systemic-risk assessment without meeting Anthropic's definition for catastrophic risk or California’s numerical harm thresholds. A global program should retain the EU’s broader inquiry instead of applying a catastrophic-harm cutoff to every risk.**

---

## (3) Transparency Has Different Audiences

The **EU’s transparency obligations serve several audiences**. Technical documentation supports supervision; information supplied to downstream providers helps them understand capabilities and limitations and meet their own obligations; the training-content summary informs the public. ([Article 53](https://ai-act-law.eu/article/53/?ref=thedelatorrereview.com); Annexes [XI](https://ai-act-law.eu/annex/11/?ref=thedelatorrereview.com)–[XII](https://ai-act-law.eu/annex/12/?ref=thedelatorrereview.com).)

**California places particular emphasis on public disclosure of safety frameworks and assessment summaries.** These public materials cannot substitute for the EU documentation supplied to regulators and downstream providers.

**Anthropic’s proposal likewise distinguishes between public disclosure and government reporting.** Covered developers would publish their safety frameworks, risk reports at least every six months, and system cards for qualifying deployments, while submitting annual compliance certifications and critical safety incident reports to a designated agency. Independent evaluators would also publish reviews of developers’ risk reports. **These disclosures focus on catastrophic-risk governance and accountability**, rather than the EU’s separate requirements for downstream-provider information and public training-content summaries. (Framework, [pp. 5–8](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com#page=5).)

---

> **A public system card can support transparency across jurisdictions, but it does not replace the EU’s requirements for technical documentation for EU regulators and information that enables downstream providers to understand the model and meet their own obligations.**

---

## (4) EU Oversight Already Includes Intervention Powers

The EU comparison also changes how *“stronger oversight”* should be understood. The Commission’s powers include requesting information, conducting model evaluations, and requiring corrective measures. [Article 93](https://ai-act-law.eu/article/93/?ref=thedelatorrereview.com) of the EU AI Act provides for restrictions on availability, withdrawal, or recall in the circumstances it specifies. (See also, [AI Office Q&A on enforcement powers](https://digital-strategy.ec.europa.eu/en/faqs/general-purpose-ai-models-ai-act-questions-answers?ref=thedelatorrereview.com))

**Anthropic’s call for deployment intervention therefore has a closer counterpart in the EU’s supervisory structure than in TFAIA’s specific civil-penalty provision.** This is a comparison of regulatory functions, not a claim that the procedures or legal tests are identical. 

---

> [**Article 55** ](https://ai-act-law.eu/article/55/?ref=thedelatorrereview.com)**of the EU AI Act does not itself impose a universal requirement that every provider retain an independent evaluator. Provider testing, independent assessment, and regulator-led evaluation are distinct concepts.**

---

## (5) The Code of Practice Is a Compliance Route

The EU’s **General-Purpose AI Code of Practice**, published on July, 2025, provides a voluntary route for demonstrating compliance with the AI Act’s binding GPAI obligations. Its [Transparency chapter](https://ec.europa.eu/newsroom/dae/redirection/document/118120?ref=thedelatorrereview.com) addresses model documentation and information for downstream providers and regulators; its [Copyright chapter](https://ec.europa.eu/newsroom/dae/redirection/document/118115?ref=thedelatorrereview.com) supports compliance with EU copyright law; and its [Safety and Security chapter](https://ec.europa.eu/newsroom/dae/redirection/document/118119?ref=thedelatorrereview.com) addresses the additional obligations for providers of GPAI models with systemic risk. Providers that do not follow the Code must demonstrate adequate alternative means of compliance. ([Commission overview of the Code](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai?ref=thedelatorrereview.com); Articles [53(4)](https://ai-act-law.eu/article/53/?ref=thedelatorrereview.com), [56](https://ai-act-law.eu/article/56/?ref=thedelatorrereview.com) of the EU AI Act.)

---

> The EU Code of Practice serves a different legal function from each of the other instruments. The **EU AI Act** establishes binding obligations; **Anthropic’s proposal** recommends requirements for future legislation; and a **large frontier developer’s framework under California’s TFAIA** sets out commitments the developer must legally follow. The **Code**, by contrast, provides a voluntary means of demonstrating compliance with the EU AI Act’s existing obligations.

---

# Federal Preemption and Societal Resilience

**Anthropic recommends preserving state authority unless Congress enacts a sufficiently strong federal regime, with preemption limited to specified governance functions.** Its resilience agenda also reaches beyond developer compliance into biological preparedness and cyber defense. ([Policy overview](https://www.anthropic.com/policy-on-the-ai-exponential?ref=thedelatorrereview.com))

---

> **For California, this raises a coordination question: which obligations would a future federal law replace, and which would remain? Similar requirements do not produce preemption on their own. That result would depend on enacted federal text.**

---

# Practical Implications for Model Governance

Developers operating across jurisdictions can organize their evidence and governance around common processes while preserving separate legal assessments.

- **First, classify coverage independently.** Distinguish California’s foundation and frontier models from the EU’s GPAI and systemic-risk categories.
- **Second, maintain a shared evidence base.** Evaluation results, mitigation records, security controls, and release decisions can support several obligations.
- **Third, separate disclosure products.** Public summaries, confidential regulatory submissions, and downstream-provider documentation serve different purposes and may require different detail.
- **Fourth, preserve distinct incident rules.** A single internal intake process can feed multiple reporting assessments, with deadlines and recipients tracked separately.
- **Finally, control changes to safety commitments.** Legal, technical, and governance teams should review material revisions together.

These approaches share a view that powerful models require governance at the model and provider levels. Their differences determine who is covered, what risks must be assessed, what evidence must be disclosed, and when public authorities can intervene. Those distinctions will matter as companies translate converging safety practices into compliance across jurisdictions.

# Relevant Resources

## Primary Materials

- [Anthropic’s Advanced AI Framework, June 2026](https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdf?ref=thedelatorrereview.com) — Proposed developer obligations and societal resilience measures.
- [Anthropic — Policy on the AI Exponential](https://www.anthropic.com/policy-on-the-ai-exponential?ref=thedelatorrereview.com) — Announcement presenting the Advanced AI Framework alongside the Economic Policy Framework. Explains Anthropic’s call for transparency, independent evaluation, security, government intervention powers, and societal resilience, as well as its position on preserving state authority.
- [California SB 53](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill%5Fid=202520260SB53&ref=thedelatorrereview.com) — Enacted legislation, including Cal. Bus. & Prof. Code §§ 22757.10–22757.16.
- [EU AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng?ref=thedelatorrereview.com) — Particularly Articles 3, 51–56, and 88–93, and Annexes XI–XIII.
- [European Commission GPAI Questions and Answers](https://digital-strategy.ec.europa.eu/en/faqs/general-purpose-ai-models-ai-act-questions-answers?ref=thedelatorrereview.com) — Scope, obligations, and supervisory powers.
- [EU GPAI Code of Practice](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai?ref=thedelatorrereview.com) — Voluntary means of demonstrating compliance with applicable GPAI obligations.

## Related Articles From The de la Torre Review

- [California’s Frontier AI Transparency Act: Building an Ongoing Safety and Oversight System for Powerful General-Purpose AI Models](https://www.thedelatorrereview.com/californias-frontier-ai-transparency-act-building-an-ongoing-safety-and-oversight-system-for-powerful-general-purpose-ai-models/) — Coverage, safety frameworks, disclosures, incident reporting, and enforcement.
- [General-Purpose AI Models Under the EU AI Act: Classification, Obligations, and Systemic Risk](https://www.thedelatorrereview.com/general-purpose-ai-models-under-the-eu-ai-act-classification-obligations-and-systemic-risk/) — GPAI classification, provider responsibilities, open-source accommodations, and systemic-risk duties.

![](https://storage.ghost.io/c/54/ef/54efeb65-4f2f-479e-b4bb-a9ee526feeaa/content/images/2026/10/ChatGPT-Image-Jul-4--2026-at-04_44_14-PM-29.png)