Legislature(2025 - 2026)BELTZ 105 (TSBldg)
02/06/2025 03:30 PM Senate STATE AFFAIRS
Note: the audio ![]()
and video ![]()
recordings are distinct records and are obtained from different sources. As such there may be key differences between the two. The audio recordings are captured by our records offices as the official record of the meeting and will have more accurate timestamps. Use the icons to switch between them.
| Audio | Topic |
|---|---|
| Start | |
| Presentation(s): Ai and State Functions | |
| Presentation(s): A.i. Deepfakes and Chatbots | |
| Adjourn |
* first hearing in first committee of referral
+ teleconferenced
= bill was previously heard/scheduled
+ teleconferenced
= bill was previously heard/scheduled
| + | TELECONFERENCED | ||
ALASKA STATE LEGISLATURE
SENATE STATE AFFAIRS STANDING COMMITTEE
February 6, 2025
3:31 p.m.
DRAFT
MEMBERS PRESENT
Senator Scott Kawasaki, Chair
Senator Jesse Bjorkman, Vice Chair
Senator Elvi Gray-Jackson
Senator Robert Yundt
MEMBERS ABSENT
Senator Bill Wielechowski
COMMITTEE CALENDAR
PRESENTATION(S): AI AND STATE FUNCTIONS
- HEARD
PRESENTATION(S): A.I. DEEPFAKES AND CHATBOTS
- HEARD
PREVIOUS COMMITTEE ACTION
No previous action to record
WITNESS REGISTER
BILL SMITH, Chief Information Officer (CIO)
Office of Information Technology (OIT)
Department of Administration
Anchorage, Alaska
POSITION STATEMENT: Provided the presentation: AI and State
Functions.
ILLANA BELLER, Organizing Manager
Public Citizen
Washington D.C.
POSITION STATEMENT: Provided the presentation: A.I. Deepfakes
and Chatbots.
ACTION NARRATIVE
3:31:22 PM
CHAIR KAWASAKI called the Senate State Affairs Standing
Committee meeting to order at 3:31 p.m. Present at the call to
order were Senators Gray-Jackson, Yundt and Chair Kawasaki.
Senator Bjorkman arrived thereafter.
^PRESENTATION(S): AI and STATE FUNCTIONS
PRESENTATION(S): AI and STATE FUNCTIONS
3:32:47 PM
CHAIR KAWASAKI announced the presentation: AI and STATE
FUNCTIONS.
3:32:59 PM
BILL SMITH, Chief Information Officer (CIO), Office of
Information Technology (OIT), Department of Administration,
Anchorage, Alaska, provided the presentation AI and State
Functions. He moved to slide 2, Generative Artificial
Intelligence (AI) and he defined generative artificial
intelligence as; creating systems that can create things that
didn't exist before. He stated that the last several years
technology has gotten to a point where it can create new things
that have not existed before, and technology is able to look at
multimodal forms of information. An example would be an image,
AI can look at the image and reference other content it has
received before; from other places and make assumptions from the
information.
3:34:49 PM
MR. SMITH continued slide 2. He stated over the past year or
two, state agencies, within the executive branch, have been
exploring how AI can meaningfully improve operations. He stated
that the focus has been on identifying real use casesstarting
with a specific problem and then finding the right tool to
address it. The Office of Information Technology (OIT) has led
efforts to educate teams on what AI can and cannot do, promote
responsible use, and help departments align the right tools with
their specific needs. He said OIT has been reviewing data systems
to ensure information is secure and AI-ready, while developing
clear guidance around the use and security of these tools. He
stated that OIT is assessing the existing enterprise tools to
see how new AI capabilities can be integrated, so OIT can make
the most of what they already have.
3:36:55 PM
CHAIR KAWASAKI announced Senator Bjorkman joined the committee.
3:37:03 PM
MR. SMITH moved to slide 3 and talked about the following
artificial intelligence tools:
[Original punctuation provided.]
Generative AI Approaches: End User AI, Developer AI,
Custom Build AI
End User AI
Off-the-shelf options with AI embedded in currently
used tools.
AI embedded into third-party Software Applications by
providers to improve functionality.
Developer AI
Custom AI configuration using existing AI tools.
State of Alaska Professional and Cloud Developer
tools-AI tools that allow for secure, custom
configuration.
Custom Build AI
Custom AI models from scratch.
Large scale implementation and creation/training of
foundational model.
MR. SMITH stated that OIT hasn't seen any big custom build
request come to them, but they are working with the review
board, so OIT is able to make sure they are asking the requester
the right questions and things are done in a responsible manner.
3:40:04 PM
CHAIR KAWASAKI mentioned that Mr. Smith referenced types of
questions to ask a custom build AI requester. He asked what are
some of the caveats OIT watches for when doing custom build AI.
3:40:31 PM
MR. SMITH responded that when OIT evaluates AI platforms,
especially custom builds, we focus on a few key items. First, is
data secure and safely used. The AI platforms must follow OIT's
security and compliance standardswhether for health, criminal
or other areas. Second, OIT looks at responsible use AI
principles like privacy, reliability, and safety to ensure the
tool fits its intended use. He said the left two columns of the
slide, End User AI and Developer AI, provide OIT with an
opportunity to move more rapidly and effectively because they
are built with data and in data environments that are secure
with the right protocol compliance in place.
3:42:04 PM
MR. SMITH moved to slide 4, Enterprise Artificial Intelligence
(AI) Tools. He explained what OIT is doing right now in the
executive branch with AI tools. He named some examples: First,
OIT already uses AI in cybersecurity systems that monitor
network traffic and flag unusual behavior for its teams to check
out. Second, OIT is about to pilot AI assistants in office
softwarelike Outlook, Word, and Excel. These tools help
summarize emails, draft messages, pull action items from meeting
transcripts, and more. It's meant to boost productivity without
requiring people to learn new tools. He stated that since
there's a cost, OIT will start small, testing the value in real
work settings. Then OIT will share what works best so leaders
can decide if and where it's worth expanding.
3:45:22 PM
SENATOR BJORKMAN stated that legislators often need to compare
two documents and analyze the differences. He said he hasn't
found a great AI tool for that yet. He asked if Mr. Smith knew
of one he could try.
3:45:44 PM
MR. SMITH answered that he doesn't know of any but will talk
with his team to see if there's an AI tool that would compare
two documents and analyze the differences.
3:46:22 PM
MR. SMITH continued slide 4. He stated that the next category is
"professional" usersthose between end users and developers.
Think of an accountant who needs to work with data but doesn't
code. He said OIT is using tools like Co-Pilot Studio to help
build assistants for these users. He noted one example: OIT
created a chatbot that understands the security policies.
Instead of digging through long documents, an employee can ask a
question like, "Can I check the weather on my work computer?"
and get a clear answer with a policy reference.
3:47:56 PM
CHAIR KAWASAKI commented on the example given and asked if OIT
are double checking the answer AI gives to match what a person
would find if they read through all the documents.
3:48:17 PM
MR. SMITH responded that OIT looks for the accurate and
controlled data. He stated that part of the testing performed by
OIT is proof of concept using a chatbot that pulls from state
websites. He stated that it worked well but sometimes gave
outdated infolike listing someone no longer in a role. That's
because it pulled from an old document and OIT had to tweak it
to prioritize newer sources. He said the tool needs both clean
data and tuning to get accurate answers. OIT tests both
thoroughly.
3:49:44 PM
SENATOR BJORKMAN mentioned complaints that using the ALVIN
system through the DMV is very slow. He asked if AI could
identify and propose solutions to fix the system like ALVIN.
3:50:29 PM
MR. SMITH answered at this time OIT doesn't have an AI tool for
that specifically but it is the type of thing AI is used for,
and the state may have it in the future. He stated that AI is
good at solving problems but not good at solving mysteries.
3:51:40 PM
MR. SMITH continued with slide 4. He explained the Developer
group needs some IT and programming skills but uses tools
already in OIT's cloud setup. He said several departments are
working with vendors on AI solutions, made possible by the
earlier Rapid Cloud Migration Project, which created secure,
compliant cloud spaces for each department. He stated that
upgrading Office licenses also enabled access to advanced AI
features. These tools run in OIT's private cloud, keeping data
secure and allowing faster more confident AI adoption to better
serve Alaskans.
3:54:15 PM
CHAIR KAWASAKI said regarding testing the system he asked
whether OIT tests for specific things like biases within the
system and does the state have a written policy.
3:54:43 PM
MR. SMITH answered that OIT doesn't currently have a specific
policy on AI bias, mainly because of the kind of AI OIT is using
right now. Most of the tools used are pre-trained models brought
into OIT's secure environment, where AI learns from our own
data. Since those models are vetted beforehand, the risk of bias
is lower than with public-facing AI. He stated that if OIT moves
toward building its own large language models, it will need to
be more careful about training and testing for bias. For now,
OIT's focus has been on educating employees about handling data
responsibly, whether they're using AI, email, or other systems.
OIT is updating their acceptable use policy to reflect that.
3:57:30 PM
CHAIR KAWASAKI asked Mr. Smith to explain front facing systems
that have an automatic chat bot that answers questions.
3:58:08 PM
MR. SMITH responded that natural language chatbots make it
easier for people to get info without needing to phrase things
perfectly. AI chatbots remember context so you can have a back-
and-forth conversation. He stated many state departments had AI
chatbot on their wish list to start implementing it. OIT has
been testing several natural language chatbots and plan to roll
them out across sites and services soon to reduce friction and
make access easier. One early version is already live on the DMV
siteit's AI-driven and getting better over time.
4:02:32 PM
At ease.
^PRESENTATION(S): A.I. DEEPFAKES and CHATBOTS
PRESENTATION(S): A.I. DEEPFAKES and CHATBOTS
4:03:02 PM
CHAIR KAWASAKI reconvened the meeting and announced the
Presentation: A.I. Deepfakes and Chatbots
4:05:41 PM
ILLANA BELLER, Organizing Manager, Public Citizen, Washington
D.C., provided a presentation A.I. Deepfakes and Chatbots.
4:06:55 PM
MS. BELLER moved to slide 3, What is a deepfake and how are they
used in an election's context. She stated that a deepfake is
content that is fabricated using technology. This can be done
through audio, video or an image. She explained it depicts
someone doing or saying something they never did or said in real
life. She restated the definition of deepfakes but in reference
to election candidates.
4:08:29 PM
MS. BELLER moved to slide 4 and continued to emphasize that
deepfakes are more prominent now than before. She stated that a
deepfake scandal occurred in Slovakia's election in 2024.
4:10:12 PM
MS. BELLER moved to slide 5. She gave another example of how
deepfakes are used in elections. She explained that during the
New Hampshire primary, residents received deepfake robocalls of
President Biden telling the voters not to vote this election.
4:11:00 PM
MS. BELLER moved to slide 6. She gave another example how
deepfakes were used during a political campaign in North
Carolina involving a Political Action Committee (PAC) and Mark
Walker. She explained the process of how the deepfake occurred
and how PAC was supporting candidate A and then PAC circulated a
deepfake of candidate B (Mark Walker). She continued with saying
the video deepfake showed candidate B telling the public he was
not qualified for office and predicting candidate A was going to
win the election.
4:12:10 PM
MS. BELLER moved to slide 7 and stated that a long-term
consequence of deepfakes is peoples distrust of media.
4:12:48 PM
MS. BELLER moved to slide 8 and explained that technology is
improving rapidly. She stated deepfakes are improving in
quality, quantity and accessibility. She emphasized how quickly
a bad deepfake can circulate and can cause an outsized effect.
4:14:39 PM
MS. BELLER moved to slide 9 and spoke about the following:
[Original punctuation provided.]
Legislation Addressing this Issue.
• 21 states have passed legislation to
regulate deepfakes.
• 49 States and D.C. have introduced
legislation
-Mostly in this session and last
-Most that did not pass was because of time.
• 18 States have already introduced this
legislation this session.
• Strong bipartisan support, unanimous votes.
4:15:52 PM
MS. BELLER moved to slide 10 and explained the United States map
shows where election deepfake bills passed or are in play as of
January 2025. She said the dark blue states passed deepfake
bills before 2024. She said the lighter blue represent states
that passed deepfake bills in 2024. She said the green
represents states that have legislation pending. She stated the
deepfake bills are passing in republican and democratic states
with an almost nearly unanimous support.
4:16:40 PM
MS. BELLER moved to slide 11 and spoke about the following:
[Original punctuation provided.]
Key Elements of Anti-Fraudulent Deepfake Legislation
• Prohibit distribution of unlabeled deepfakes -
Why disclosure not ban?
She stated there are a couple bills that are a full ban. She
recommends people should require a disclosure instead of a
complete ban, it's less likely to be challenged by the first
amendment.
• Standards for disclosure - must be very clear and
prominent
• Covers all people - not just candidates, parties
and committees. e.g. influencers
• Usually within a certain number of days of E-day
• Only covers people who knowingly circulate a
deepfake
• Establish a right for affected parties to seek
injunction to take it down
• Establishes penalties
4:19:44 PM
MS. BELLER moved to slide 12 and spoke about the following:
[Original punctuation provided.]
Protections
• No liability for broadcasters or platforms that
make reasonable effort to prevent deepfakes, or
that show deepfakes as part of news coverage and
describe as deepfakes
• Satire and parody are protected
• Individuals who are unknowingly reposting are not
held liable
4:20:59 PM
MS. BELLER moved to slide 13 and spoke about the following:
[Original punctuation provided.]
Lessons Learned
- Clear definitions is critical
-Depicts someone doing or saying something they never
did or said
-Provides a fundamentally different understanding of
the persons speech or behavior
-Intent to undermine candidates' reputation or
otherwise deceive voters
-Satire and parody exemptions
-Liability must be on the distributor
4:23:54 PM
MS. BELLER moved to slide 14 and 15, Non-Consensual Intimate
Deepfakes and What is an intimate deepfake. She said an intimate
deepfake is fabricated content of a sexual act or in the nude,
in the form of a video or image that's not real. She stated the
intimate deepfakes are more common than election deepfakes.
4:25:08 PM
MS. BELLER moved to slide 16 and spoke about the following:
[Original punctuation provided.]
Pervasiveness of this Problem
• Intimate deepfakes make up 98 percent of all
deepfakes
• 99 percent of victims are women
• Children (girls) are a significant percentage of
the victims
• A study recently found that 57 percent of those
under 18 years of age are concerned about
becoming victims to intimate deepfakes and 10
percent of individuals reported being a victim of
intimate deepfakes, knowing a victim, or both.
• WIRED found on Telegram at least 50 bots that
claim to create explicit photos or videos of
people with only a couple of clicks - these bots
have over 4 million monthly users
4:26:45 PM
MS. BELLER moved to slide 17, Examples. She mentioned the
examples of intimate deepfakes around the world reported by the
news.
4:28:15 PM
MS. BELLER moved to slide 18. She read the different types of
harm that can be the outcome of being a victim of intimate
deepfakes. She said deepfake pornography is being used for
sexual extortion, as a result there have been a few cases of
suicide.
4:29:41 PM
MS. BELLER moved to slide 19, State Tracker. She discussed the
United States map that showed the 30 dark green states that have
passed legislation, the light green states have legislation
that's pending, the bright green states have passed legislation,
but it only covers minors. 4 out of 5 states that passed the
child only deepfake laws last year; have now introduced laws
that would cover everyone.
4:30:54 PM
MS. BELLER moved to slide 20 and spoke about the following:
[Original punctuation provided.]
Legislation Passed in the States
• 30 states have enacted legislation + 1 more just
passed
• 45 states have introduced legislation
• Broad bipartisan support
4:31:20 PM
MS. BELLER moved to slide 21, Model Legislation. She stated that
Public Citizen has shared a model bill and have seen states take
a few main approaches: about half passed standalone laws, others
updated private image or revenge porn laws, and some amended
child porn laws when minors are involved. She said for states
updating private image laws, it's critical to include "intent"
in the criminal section. She stated that's because courts may
argue that privacy violations don't apply if the image isn't
technically "real," even if it looks just like the person. She
emphasized just adding "deepfake" to the definition isn't enough.
Intent to harass, threaten, or alarm usually needs to be clearly
included. She then spoke about the following:
[Original punctuation provided.]
Model Legislation.
A few important recommendations to keep in mind when
drafting this legislation:
1.Provide for both civil liability and criminal
penalties. Civil liability can afford the victim
injunctive and economic relief while the criminal
penalty can act as a deterrent.
2. Allow for a defense of consent but require more
than mere assertions of oral consent to satisfy it.
3. Disclaimers of inauthenticity or unauthorized
creation should not be permitted as a defense against
intimate deepfakes, as such disclosure does not
mitigate the reputation and psychological harm done to
the victims.
https://www.citizen.org/article/public-citizen-model-
state-law-regulating-non-consensual-intimate-
deepfakes/
4:34:58 PM
MS. BELLER moved to slide 23. Consumer Protection + Chatbot
Labeling. She stated that more people are catching on to
chatbots, and some states are starting to require clear
disclosure when someone is interacting with one. This applies to
both sales bots and bots used in areas like healthcare or legal
services. She said the concern is that bots can access personal
data and use algorithms to manipulate behaviorlike fast food
bots upselling based on past orders and people could think
they're talking to a doctor or lawyer when it's a bot. She
opined it's a simple fixjust make sure users know they're
dealing with AI. A few states have passed or introduced laws on
this, and it has bipartisan support.
4:38:05 PM
CHAIR KAWASAKI asked how to tell the difference between harmless
altered deepfakes and deepfakes that cross a legal and ethical
line.
4:38:45 PM
MS. BELLER answered that Public Citizen's model defines
synthetic media as manipulated images, audio, or video showing
someone's appearance, speech, or conduct. When it comes to
deceptive and fraudulent deepfakes are what Public Citizen is
seeking to regulate. This type of deepfake is synthetic media
that depicts a candidate with the intent to injure their
reputation or deceive the voter or mislead voters. She stated
that the deepfake must appear real to a reasonable person and
show someone saying or doing something they didn't. The key is
intent and whether it gives a viewer a fundamentally false
impression of what happened. She said minor edits, like removing
a mole or adjusting height, wouldn't meet that threshold.
4:41:31 PM
CHAIR KAWASAKI asked if child pornography should be illegal
regardless how its presented, by artificial intelligence or not.
4:42:12 PM
MS. BELLER responded that many states are updating existing
child pornography laws to include synthetic media and deepfakes.
Most of these updates expand the definition to cover not just
distribution but also possessionthough not all states include
possession. She stated that it's up to legislators, but the
trend is to tweak current laws to fully cover this content.
4:44:54 PM
There being no further business to come before the committee,
Chair Kawasaki adjourned the Senate State Affairs Standing
Committee meeting at 4:44 p.m.
| Document Name | Date/Time | Subjects |
|---|---|---|
| SSTA DOA OIT AI Presentation 2.6.25.pdf |
SSTA 2/6/2025 3:30:00 PM |
|
| AI Presentation - AK Senate State Affairs Committee.pdf |
SSTA 2/6/2025 3:30:00 PM |