,i  #aah? ,,congress  #b;d ,session  ,h4
 ,r4 #idff  ,,9 ,,! ,,h\se ,,(          
,,repres5tatives  ,septemb] #f1         
#bjbd,mr4 ,baird 7= himself & ,mr4      
,lieu7 introduc$ ! foll[+ bill2 :i* 0   
ref]r$ to ! ,committee on ,sci5ce1      
,space1 & ,te*nology  ,a ,,bill  ,to    
direct ! ,national ,in/itute ( ,/&>ds & 
,te*nology to catalog & evaluate em]g+  
practices & norms = communicat+ c]ta9   
*>act]i/ics ( artificial intellig5ce    
sy/ems1 includ+ relat+ to transp>5cy1   
robu/ness1 resili5ce1 security1 safety1 
& usability1 & = o!r purposes4  #a4     
,%ort title ,?is ,act may be cit$ z !   
,,ai ,developm5t ,practices ,act (      
#bjbd4  #b4 ,,ni/ rese>* on developm5t  
be/ practices ,section #bb;,a ( !       
,national ,in/itute ( ,/&>ds &          
,te*nology ,act 7#ae ,u4,s4,c4          
#bgh;h-#a7 is am5d$-- 7#a7 by r$esignat+
 subsection 7h7 z subsection 7i72 &     
7#b7 by ins]t+ aft] subsection 7g7 !    
foll[+ new subsection3  7h7 ,assessm5t (
 ! practices ( artificial intellig5ce   

developm5t  7#a7 ,9 g5]al ,! ,director (
 ! ,national ,in/itute ( ,/&>ds &       
,te*nology 79 ?is subsection ref]r$ to z
 ! ,director7 %all1 subject to !        
availability ( appropriations1 develop1 
& p]iodically update1 9 collaboration ) 
o!r public & private sector             
organizations1 volunt>y guidance =      
practices & guidel9es relat+ to !       
developm5t1 release1 & assessm5t (      
artificial intellig5ce sy/ems4 ,su*     
guidel9es %all satisfy ! foll[+3 7,a7   
,def9e me?ods & guidel9es = develop+    
reasonable risk tol]ances = v>i\s use   
cases ( artificial intellig5ce sy/ems   
bas$ on ! foll[+3 7i7 ,! risks associat$
 ) ! int5d$ & un9t5d$ applications1 use 
cases1 & outcomes ( ! artificial        
intellig5ce sy/em at issue1 bas$ on !   
guidel9es specifi$ 9 ! volunt>y risk    
managem5t framework = tru/wor?y         
artificial intellig5ce sy/ems1 or       
successor framework1 au?oriz$ "u        
subsection 7c71 :i* may include diff]5t 
categories ( risk1 su* z ! foll[+3 7,i7 

,security risks1 includ+ ?reats to      
national security4  7,,ii7 ,economic    
risks1 includ+ ?reats to economic       
opportunities4  7,,iii7 ,social risks1  
includ+ infr+em5t ^u con/itutional      
rights1 privileges1 or lib]ties4  7ii7  
,su* o!r factors z ! ,director det]m9es 
appropriate & consi/5t ) ?is subsection4
  7,b7 ,categorize & li/ practices &    
norms = communicat+ relevant            
*>act]i/ics1 includ+ robu/ness1         
resili5ce1 security1 safety1 fairness1  
privacy1 validation1 reliability1       
acc\ntability1 & usability1 ( artificial
 intellig5ce sy/ems1 & includ+ any      
*>act]i/ics id5tifi$ by ! volunt>y risk 
managem5t framework = tru/wor?y         
artificial intellig5ce sy/ems1 or       
successor framework1 au?oriz$ "u        
subsection 7c74 ,su* practices & norms  
may relate to ! foll[+3 7i7             
,docum5tation ( tra9+ & evaluation      
datasets1 su* z in=mation & /ati/ics    
ab\t a dataset's size1 curation1        
annotation1 & s\rces1 & ! protocols = a 

dataset's selection1 creators1          
prov5ance1 process+1 augm5tation1       
filt]s1 inclusion ( p]sonally           
id5tifiable in=mation1 & intellectual   
prop]ty usage4  7ii7 ,docum5tation (    
model in=mation1 su* z a model's        
developm5t /ages1 tra9+ objectives1     
tra9+ /rategies1 inf]5ce objectives1    
capabilities1 reproducibility (         
capabilities1 input & output modalities1
 compon5ts1 size1 & ar*itecture4  7iii7 
,evaluation ( b5*m>ks = multi-metric    
assessm5ts1 su* z an assessm5t ( an     
appropriate comb9ation ( robu/ness1     
resili5ce1 security1 safety1 fairness1  
privacy1 accuracy1 validity1            
reliability1 acc\ntability1 usability1  
transp>5cy1 effici5cy1 & calibration1 & 
any *>act]i/ics id5tifi$ by ! volunt>y  
risk managem5t framework = tru/wor?y    
artificial intellig5ce sy/ems1 or       
successor framework1 au?oriz$ "u        
subsection 7c74  7iv7 ,metrics &        
me?odologies = evaluations ( artificial 
intellig5ce sy/ems1 su* z e/abli%+      

evaluation datasets4  7v7 ,public       
report+ ( artificial intellig5ce sy/ems'
 capabilities1 limitations1 & possible  
areas ( appropriate & inappropriate use4
  7vi7 ,disclosure ( security practices1
 su* z artificial intellig5ce r$ team+ &
 ?ird-p>ty assessm5ts1 t 7 us$ 9 !      
developm5t ( an artificial intellig5ce  
sy/em4  7vii7 ,h[ to release to ! public
 compon5ts ( an artificial intellig5ce  
sy/em or in=mation ab\t an artificial   
intellig5ce sy/em1 includ+ aspects ( !  
model1 associat$ tra9+ data1 & lic5se   
agreem5ts4  7viii7 ,approa*es & *annels 
= collaboration & k-%>+ ( be/ practices 
across indu/ry1 gov]nm5ts1 civil        
society1 & academia4  7ix7 ,su* o!r     
categories z ! ,director det]m9es       
appropriate & consi/5t ) ?is subsection4
  7,c7 ,= ea* practice & norm categoriz$
 & li/$ 9 accordance ) subp>agraph 7,b71
 provide recomm5dations & practices =   
utiliz+ su* practice or norm4  7#b7     
,implem5tation ,9 conduct+ ! ,director's
 duties "u p>agraph 7#a71 ! ,director   

%all c>ry out ! foll[+3 7,a7 ,update !  
volunt>y risk managem5t framework =     
tru/wor?y artificial intellig5ce sy/ems1
 or successor framework1 au?oriz$ "u    
subsection 7c7 z ! ,director det]m9es   
appropriate4  7,b7 ,ensure t volunt>y   
guidance develop$ 9 p>agraph 7#a7 is    
bas$ on int]national /&>ds & indu/ry be/
 practices to ! ext5t possible &        
practical4  7,c7 ,n prescribe or o!rwise
 require ! use ( specific in=mation or  
communications te*nology products or    
s]vices4  7,d7 ,collaborate ) public1   
indu/ry1 & academic entities z !        
,director det]m9es appropriate1 includ+ 
conduct+ p]iodic outrea* to receive     
public input f public1 indu/ry1 &       
academic /akehold]s4  7#c7 ,report ,9   
conduct+ ! ,director's duties "u        
p>agraph 7#a71 ! ,director %all1 n lat] 
?an #ah mon?s aft] ! date ( ! enactm5t (
 ?is subsection1 brief ! ,committee on  
,sci5ce1 ,space1 & ,te*nology ( ! ,h\se 
( ,repres5tatives & ! ,committee on     
,comm]ce1 ,sci5ce1 & ,transportation ( !

 ,s5ate on ! foll[+3 7,a7 ,new or updat$
 mat]ials1 programs1 or sy/ems t 7      
produc$ z a result ( c>ry+ out ?is      
subsection4  7,b7 ,policy recomm5dations
 ( ! ,director t c\ld facilitate &      
improve communication & coord9ation     
betwe5 ! private sector & relevant      
,f$]al ag5cies reg>d+ implem5t+ !       
recomm5d$ practices id5tifi$ 9 ?is      
subsection4  7#d7 ,def9itions ,9 ?is    
subsection3 7,a7 ,artificial intellig5ce
 r$ team+ ,! t]m artificial intellig5ce 
r$ team+ means a /ructur$ te/+ (        
adv]s>ial ef=ts to f9d flaws &          
vuln]abilities 9 an artificial          
intellig5ce sy/em & id5tify risks1      
flaws1 & vuln]abilities ( artificial    
intellig5ce sy/ems1 su* z h>mful outputs
 f su* sy/em1 un=ese5 or undesirable    
sy/em behaviors1 limitations1 & pot5tial
 risks associat$ ) ! misuse ( su* sy/em4
  7,b7 ,artificial intellig5ce sy/em ,! 
t]m artificial intellig5ce sy/em has !  
mean+ giv5 su* t]m 9 section #gbbc ( !  
,advanc+ ,am]ican ,,ai ,act 7#dj        

,u4,s4,c4 #aacja note2 z enact$ z "p (  
title ,,lxxii ( division ,g ( ! ,james  
,m4 ,inh(e ,national ,def5se            
,au?orization ,act = ,fiscal ,ye> #bjbc2
 ,public ,law #aag-bfc74  4