Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence

What does Executive Order 14110 do?

Commerce and NIST had testing and risk guidance, while OMB and agencies were assigned governance and inventory duties with staged deadlines.

Synthetic content: a report, then two different guidance steps

Section 4.5 did not itself require every private AI image or text output to carry a watermark. It first directed Commerce to survey technical approaches, then develop guidance with OMB coordination. OMB’s later step concerned labeling and authenticating digital content that federal agencies produce or publish. The FAR Council was to consider appropriate amendments, not automatically insert a new contract clause when the President signed the order. Original section 4.5(a)–(d).

The first clock was 240 days from October 30, 2023, which is June 26, 2024 on a calendar-day count. Commerce’s next 180-day clock ran from report submission; OMB’s separate 180-day clock ran from development of Commerce’s guidance. The public release date of a technical report does not, without additional evidence, establish the internal submission date or the date that starts each downstream clock. These historical instructions ended as commands under this order when EO 14110 was revoked; an independently operative instrument requires its own authority. Original timing triggers.

NIST’s record distinguishes draft review from final technical publication. Its comments page says draft materials were posted April 29, 2024; the May 7 Federal Register notice formally sought comments on four drafts, including AI 100-4, through June 2. On November 20, NIST published the final Reducing Risks Posed by Synthetic Content report and described its approaches as voluntary. Those records establish public technical work, not that every later OMB or FAR step was completed. Draft posting record; Comment notice; Final publication; Release and voluntary scope.

The report’s summary also gives an important practical limit: information about a file’s origin or editing history is not a guarantee that its message is true or trustworthy. A genuine item can be presented out of context, and no one transparency technique is a complete solution. Accordingly, a watermark, provenance record, detector result and legal authentication requirement should not be treated as interchangeable. NIST AI 100-4, summary and introduction, pages 1–3.

Research training, IP enforcement and competition were different tracks

Section 5.2(b) gave Energy 120 days to establish a scientific-training pilot in coordination with NSF; training 500 new researchers by 2025 was its goal. DOE’s public workforce portal describes pilot opportunities and repeats the prospective training target. That is evidence of a public implementation resource, not a certified count proving that the goal was met. It is also distinct from the NSF-led NAIRR pilot already described above. Original section 5.2(b); DOE and NSF workforce resource.

Section 5.2(d) separately addressed AI-related intellectual-property risks through DHS’s IPR Center, with Attorney General consultation. The program combined analysis, appropriate enforcement, information sharing, private-sector guidance and assistance to the Intellectual Property Enforcement Coordinator. FBI and CBP were named coordination partners, not newly assigned to decide every copyright dispute. The provision did not itself adjudicate whether a particular model’s training infringed a particular owner’s rights. Exact program duties, section 5.2(d)(i)–(v).

Competition instructions also differed in force. Agencies developing AI policy were directed to use their existing lawful powers appropriately; the FTC was encouraged to consider its authorities. Commerce’s semiconductor directions concerned implementation of the CHIPS Act, including flexible membership and access for startups and small firms. They were not a new appropriation enacted by executive order. The original NIST-hosted consortium fact sheet documents proposed investments and an initial pre-membership input program, rather than proof that all anticipated funding was spent. Section 5.3(a)–(b); Original consortium and planned-investment record.

The semiconductor implementation later changed. In its August 6, 2026 audit, GAO reported that Commerce canceled Natcast’s award in August 2025 and that Natcast ceased operations the following month. GAO described a new April 2026 charter and identified gaps in Commerce’s plan to satisfy statutory NSTC requirements. Its account of Commerce’s view about Natcast’s legal establishment is an attributed executive-branch position, not a court judgment supplied by this guide. Neither canceling an award nor revoking EO 14110 repeals the separate semiconductor statute. GAO-26-109121, NSTC findings and recommendations.

Health grants, veterans’ competitions and energy research had separate tests of delivery

Section 5.2(e) directed HHS to identify and, where appropriate, prioritize responsible health-AI awards. Its examples covered personalized immune-response tools, the 2024 LEAP health-data-quality awards and NIH’s AIM-AHEAD program. These were funding priorities under existing law, not a guarantee that any applicant would receive money or that a tool would be clinically approved. The official 2024 LEAP project record identifies Columbia’s nursing-data-quality work and an Oregon Health & Science University behavioral-health information-exchange project. Its estimated 2024–2026 dates do not certify completed projects. Original health award directions; Official 2024 project descriptions and goals.

For VA, the signed text required two three-month nationwide competitions within 365 days and support for participants, including potential contract opportunities. VA announced its burnout-focused competition on October 31, 2023, then named 25 finalists on March 13, 2024. The two tracks addressed ambient clinical dictation and processing outside medical records. This shows a documented competition pathway, not proof that every finalist received a contract, that all products were deployed nationwide or that measured patient outcomes improved. Section 5.2(f); VA launch; VA finalists and tracks.

Energy’s 180-day package was broader than publishing a report. It also covered scientific-model tools, climate-risk collaboration, computing partnerships and a coordinating office. On April 29, 2024, DOE announced its report on planning, permitting, operations/reliability and resilience, and a website of DOE-developed tools and models. It described VoltAIc and the PolicyAI test bed as development efforts to assist siting and environmental reviews. An analytical tool did not replace legally required review or by itself authorize construction, and the announcement does not prove every intended deployment was completed. Section 5.2(g); DOE report and opportunity areas; DOE implementation announcement.

PCAST’s April 2024 report expressly answered the order’s research charge. Its recommendations included wider access to AI resources and protected datasets, cross-sector collaboration, responsible research practices and human-AI workflows. It warned that inaccurate or biased results and weak reproducibility require attention from the start of a project. A delivered advisory report is evidence of a reporting response, not evidence that Congress funded every recommendation or that every agency adopted it. EO 14110’s subsequent revocation, already documented below, must remain separate from the history of these publications. PCAST report, letter and executive summary.

Worker guidance and justice advice were not new statutes

Section 6 separated reports about AI’s labor-market effects from employer best practices and guidance on compensation for time worked. Other agencies were to consider encouraging Labor’s principles after they were developed; the order did not give that follow-up a numerical deadline or make every principle a binding workplace rule. NSF’s education/workforce direction concerned prioritizing available resources within existing programs. Original section 6.

Labor’s official bulletin index records FAB 2024-1, on workplace AI and federal labor standards, as issued on April 29, 2024. Its current withdrawal note expressly connects that bulletin to EO 14110 and the order’s January 20, 2025 revocation by EO 14148. Readers should not treat the withdrawn bulletin as current agency guidance. The index does not supply a separate withdrawal date. Withdrawal of that document is not repeal of the Fair Labor Standards Act itself; the statutory provisions remain a separate source of legal obligations, subject to their coverage and exceptions. Labor’s issuance entry and withdrawal note; Separate federal minimum-wage statute.

The justice reporting sequence also matters. DOJ’s final report is dated December 3, 2024, but page 5 states that an Executive Report was submitted on October 29, 2024. Those are two different stages, not a reason to infer that the initial report was late merely from the final report’s cover date. The final report discusses identification/surveillance, forensics, predictive policing and risk assessment. Its recommendations emphasize testing, ongoing evaluation and human decision-making; its legal disclaimer says they are nonbinding and create no enforceable legal right. A recommendation in this report is therefore not a new court holding, a universal ban on police technology or proof that every agency adopted the recommended safeguards. DOJ report, pages 3–5 and 70–77.

Several civil-rights provisions were qualified rather than absolute commands. FHFA and CFPB were encouraged to consider using existing powers on underwriting and valuation; the Access Board was encouraged to engage on biometric risks and accessibility. Section 7.2(a) separately addressed AI in federal benefits/program administration and expressly excluded civil and criminal enforcement authorities from that subsection. These distinctions prevent a policy invitation from being mistaken for an immediately binding rule on every private business. Sections 7.2(a), 7.3(b)–(d) and 8(a).

Infrastructure assessments, guidelines and an advisory board were separate steps

Section 4.3(a) first required sector agencies and relevant regulators to assess AI risks with CISA within 90 days and at least annually afterward. Homeland Security’s guideline task had a separate 180-day clock. The later White House/OMB coordination step ran for 240 days after those guidelines were completed, not after signing. A direction to develop and take steps toward lawful mandates did not itself turn every recommendation into a final regulation on infrastructure operators. Section 4.3(a)(i)–(iv).

GAO’s December 18, 2024 audit found that agencies submitted the initial assessments, but none fully addressed all six assessment activities it examined. It reviewed 17 assessments covering 16 sectors and one subsector—not 17 separate sectors. The audit identified gaps in evaluating likelihood alongside impact and connecting mitigations to risks. It also documented the April 26, 2024 safety guidelines, an August template update and a distinct November roles-and-responsibilities framework. These historical findings are not a claim that every deficiency remained unresolved in later years. GAO-25-107435, findings and guidance history.

The AI Safety and Security Board was an advisory body, not an infrastructure licensing authority. The April 29, 2024 establishment notice expressly limited it to advice, information and recommendations. A later notice, published March 13, 2025, terminated it effective March 7, 2025 and cited EO 14217. That effective date is distinct from the notice’s publication and from EO 14110’s own January 20 revocation. Ending the Board is not, by itself, repeal of every separate infrastructure statute or an adjudication of every earlier agency rule. Official establishment notice; Official termination notice and effective date.

GAO’s later recommendation tracker supplies a more recent checkpoint. It lists the recommendation as open and attributes the pause in NSM-22 implementation to DHS’s response to EO 14239. Its July 2026 update gives March 31, 2027 as the estimated completion date. That is an estimate for addressing the recommendation, not evidence that the work was finished or a new deadline written into EO 14110. GAO recommendation status and July 2026 update.

Immigration directions did not guarantee visas or erase every residence requirement

Section 5.1 used different verbs for different tasks. State was to consider specified Skills List and domestic-renewal changes; DHS was to continue H-1B rulemaking and review appropriate pathways. Labor had a 45-day direction to request information on potential Schedule A updates. DHS also had to publish an immigration-options guide and a report covering relevant use of the system through fiscal year 2023. None of those directions automatically granted a visa or permanent residence to every AI researcher. Original section 5.1(a)–(g).

Labor’s December 21, 2023 Federal Register notice sought evidence and methods for considering Schedule A occupation-list revisions. Its original February 20, 2024 comment deadline was extended by a February 15 notice to May 13, 2024. These are consultation records, not a final decision that all AI/STEM occupations qualified for the list. The publication date should also not be confused with the order’s 45-day instruction or an earlier agency announcement. Original information request; Comment-period extension.

A different change did become an official designation: State’s revised Exchange Visitor Skills List applied on December 9, 2024. The notice changed which countries were designated, not the skill categories retained for listed countries. State’s explanation says removal of a person’s country can remove the Skills List basis for the two-year home-country physical presence requirement, including for certain earlier J-status admissions. Government funding or U.S. graduate medical education/training can independently keep that requirement applicable. The current official Skills List page explains those distinctions; a definitive individual determination should come through the official process rather than an assumption based on EO 14110. Signed-agency designation notice; State’s explanation; Official current Skills List guidance.

DHS also published a final H-1B modernization rule on December 18, 2024, with a January 17, 2025 effective date. That was a distinct regulatory instrument, not just a promise to consider rulemaking. Its historical publication does not establish that every provision remains unchanged after later immigration actions or litigation, nor does this guide determine anyone’s eligibility. The order’s revocation and the operation of later regulations must be checked separately. Final rule heading, summary and effective date.

AI voice calls: the later FCC action used the TCPA

Section 8(e) encouraged FCC consideration of AI’s effects on spectrum, networks and unwanted calls; it did not itself enact a new telephone-call prohibition. FCC’s February 8, 2024 announcement explained that its declaratory ruling treated AI-generated voices as artificial voices under the existing Telephone Consumer Protection Act. Consent requirements and applicable emergency purposes/exemptions matter: this was not a ban on every use of synthetic speech. The agency’s release describes prior express written consent for telemarketing and explains separate civil enforcement and private/state enforcement avenues. The ruling, not the press-release headline, is the operative agency instrument, and its application to any particular call requires the relevant statutory/rule conditions. Original section 8(e); FCC announcement and legal context; FCC 24-17 declaratory ruling.

The biosecurity work had several different triggers. Section 4.4(a) required a DHS risk assessment and a Defense contract for an independent study; it did not require that contracted study to be finished within the contracting deadline. Section 4.4(b) then linked research-funding requirements and DHS screening evaluation to establishment of the OSTP framework. Those later clocks should not all be calculated from the order’s signing date. Signed sections 4.4(a)–(b).

Later policy also matters. EO 14292, signed May 5, 2025, directed OSTP to revise or replace the 2024 screening framework within 90 days and directed life-science funding agencies to use providers adhering to the updated framework. NIST separately reports starting monthly baseline screening tests in August 2025, with its project still ongoing in the October 1, 2026 update. That is evidence of continuing screening work, not revival of the revoked EO 14110, proof of universal supplier compliance or confirmation that every later replacement document was issued on time. EO 14292, section 4(b); NIST’s dated program record.

For federal purchasing, the original order separated general AI-use guidance from the later work needed to implement it. The OMB progress method, NIST support tools and initial contract-alignment mechanism were due 60, 90 and 180 days after the section 10.1(b) guidance, respectively. Section 10.1(i) excluded AI used as a component of a national security system from the initial section 10.1(a)–(h) instruments; that work was addressed through the separate national-security memorandum route. Signed sections 10.1(c)–(i).

Later acquisition policy has its own scope and clocks. M-25-22 rescinded and replaced M-24-18 on April 3, 2025, and identifies statutory authority as well as later executive policy. Its future-contract rule covers solicitations and contract renewal or extension options at the specified 180-day point, while its agency-procedure review uses a separate 270-day deadline. The memorandum also excludes national security systems and intelligence-community elements and does not cover every incidental contractor use of AI. Those distinctions prevent the revoked 2023 order from being presented as the current procurement rule. M-25-22, sections 1–3.

FedRAMP’s own record distinguishes a later 2025 AI prioritization initiative from open-ended access to expedited review: it says the initiative ran from August 2025 to April 2026 and is no longer open to new entrants. Its listed early-2026 certifications should not be treated as blanket agency permission to operate every AI service or as completion of every obligation in EO 14110. FedRAMP’s current historical criteria and initiative record.

OPM’s December 29, 2023 memorandum is a concrete hiring action, not just a promised review. It authorized direct hiring for four specified AI-related occupational series at GS-9 through GS-15, while retaining public notice, qualification, priority-selection and background-investigation requirements. The same memorandum separately authorized temporary Schedule A support appointments below the Senior Executive Service; that route was not for the technical AI positions covered by direct-hire authority. OPM memorandum, pages 2–5.

The direct-hire authorization’s original endpoint was December 31, 2028, or earlier OPM termination. OPM’s current government-wide list still identifies that AI authority as GW-009 with the same conditional endpoint. This is separate evidence about an OPM instrument; it does not reinstate EO 14110, guarantee a vacancy or establish that a particular agency may hire without complying with all other applicable restrictions. Applicants should use an actual vacancy announcement and the agency’s hiring contact rather than treating the old order as an application route. OPM’s government-wide authority list and applicant guidance.

The international standards and infrastructure plans had separate follow-up clocks. Commerce’s standards engagement plan and Homeland Security’s multilateral infrastructure plan were each due within 270 days of signing, but each subsequent report was due 180 days after its respective plan was established. Publishing a standards plan is not proof that the separate infrastructure plan or either follow-up report was completed. The State-led engagement provisions also concern collaboration, common principles and voluntary commitments, not an executive order directly rewriting another country’s law. Signed sections 11(a), 11(b) and 11(d).

The order also established two differently led coordination structures. OMB chaired the section 10.1(a) interagency council for agency AI use outside national security systems, with OSTP as vice chair. The Deputy Chief of Staff for Policy chaired the section 12 White House AI Council for broader policy coordination and implementation. Keeping their chairs and functions separate avoids attributing one council’s duties or deadlines to the other. Neither a council’s creation nor its membership list proves that every resulting agency task was delivered. Signed sections 10.1(a) and 12(a)–(d).