Fable 5: Has weaponised artificial intelligence crossed the Rubicon?

In Ro­man his­to­ry, Julius Cae­sar was or­dered by the de­mo­c­ra­t­ic Sen­ate of the Ro­man Re­pub­lic not to cross the Ru­bi­con with his army. The Ru­bi­con was a shal­low riv­er in north­ern Italy that marked the bound­ary be­tween the province of Gaul and Italy. Un­der Ro­man law, gen­er­als were for­bid­den from cross­ing the riv­er in­to Italy with their armies, as do­ing so was con­sid­ered a de­c­la­ra­tion of war against the Ro­man Re­pub­lic.

In 49 BC, Cae­sar de­fied the Ro­man gov­ern­ment and crossed the Ru­bi­con, prompt­ing the Ro­man Sen­ate to flee to Greece. The move gave Cae­sar con­trol of Rome and sparked the civ­il war that trans­formed the Ro­man Re­pub­lic in­to the Ro­man Em­pire.

The phrase “cross­ing the Ru­bi­con” has since come to sig­ni­fy a piv­otal point of no re­turn—a tip­ping point in his­to­ry that re­shapes po­lit­i­cal and mil­i­tary pow­er and cre­ates a new nar­ra­tive.

In 49 BC, Cae­sar’s army dom­i­nat­ed much of Eu­rope be­cause of its su­pe­ri­or steel gla­d­ius swords and mil­i­tary tac­tics. By con­trast, mil­i­tary dom­i­nance in 2026 re­quires su­pe­ri­or­i­ty not on­ly on the bat­tle­field but al­so in cy­ber­space and dig­i­tal in­fra­struc­ture to man­age sol­diers, equip­ment and op­er­a­tions.

I in­ter­pret the mod­ern-day “cross­ing of the Ru­bi­con” as the point of no re­turn at which democ­ra­cies are threat­ened by the use of ar­ti­fi­cial in­tel­li­gence to pro­duce weapons of ter­ror or un­der­mine de­mo­c­ra­t­ic in­sti­tu­tions, much as Rome’s democ­ra­cy was de­stroyed af­ter Cae­sar crossed the Ru­bi­con.

In my view, AI has now crossed that metaphor­i­cal Ru­bi­con be­cause of its po­ten­tial to be­come a weapon in the wrong hands. Ma­jor mil­i­tary pow­ers such as the Unit­ed States and Chi­na al­ready use AI de­vel­oped for their own strate­gic pur­pos­es, al­though ac­cess to those sys­tems has gen­er­al­ly re­mained tight­ly con­trolled.

AI may have crossed the metaphor­i­cal Ru­bi­con with the pub­lic re­lease of An­throp­ic’s Fa­ble 5 mod­el. What makes Fa­ble 5 sig­nif­i­cant is that it is re­port­ed­ly de­rived from the Mythos mod­el used by the US mil­i­tary and the NSA for of­fen­sive and de­fen­sive cy­ber war­fare op­er­a­tions.

An­throp­ic re­leased Fa­ble 5 on June 9, ac­knowl­edg­ing from the out­set that the tech­nol­o­gy could cause re­al harm if mis­used. It did not take long for those con­cerns to be test­ed.

With­in days, one re­searcher re­port­ed­ly by­passed the mod­el’s safe­ty con­trols, while re­searchers at Ama­zon un­cov­ered prompts ca­pa­ble of ex­tract­ing in­for­ma­tion that could po­ten­tial­ly aid cy­ber­at­tacks.

Wash­ing­ton re­spond­ed swift­ly. On June 12, the Unit­ed States or­dered An­throp­ic to sus­pend for­eign ac­cess to Fa­ble 5. Un­able to ver­i­fy users’ iden­ti­ties in re­al time, the com­pa­ny tem­porar­i­ly shut the mod­el down for every­one.

The re­stric­tions were lift­ed on June 30, and Fa­ble 5 re­turned on Ju­ly 1 with stronger safe­guards.

It is im­por­tant to note that Fa­ble 5 was nev­er shown to have car­ried out a cy­ber­at­tack. How­ev­er, giv­en that the US mil­i­tary re­port­ed­ly us­es a sim­i­lar mod­el, Mythos, for cy­ber war­fare, au­thor­i­ties con­sid­ered the re­stric­tions nec­es­sary to mit­i­gate po­ten­tial na­tion­al se­cu­ri­ty risks.

Fa­ble 5 re­port­ed­ly out­per­forms Ope­nAI’s Chat­G­PT 5.5 and Google’s Gem­i­ni 3.1 on com­plex tasks. Its abil­i­ty to write com­put­er code and iden­ti­fy soft­ware vul­ner­a­bil­i­ties has made it a po­ten­tial tool for cy­ber­at­tacks.

The threat is not con­fined to cy­ber­space. In the­o­ry, the Mythos mod­el could al­so be used to syn­the­sise re­search on chem­i­cal or bi­o­log­i­cal weapons, po­ten­tial­ly help­ing a well-re­sourced or­gan­i­sa­tion de­vel­op weapons of mass de­struc­tion us­ing in­for­ma­tion al­ready avail­able on the in­ter­net. For­tu­nate­ly, Fa­ble 5 in­cludes safe­guards de­signed to pre­vent such mis­use.

Even with those guardrails, Fa­ble 5’s ad­vanced cod­ing ca­pa­bil­i­ties could still pro­vide lo­gis­ti­cal sup­port to mil­i­taries and non-state ac­tors. Be­yond weapons de­vel­op­ment, armed forces de­pend heav­i­ly on lo­gis­tics, man­age­ment sys­tems and IT in­fra­struc­ture.

His­tor­i­cal­ly, the first large-scale mil­i­tary de­ploy­ment of ar­ti­fi­cial in­tel­li­gence oc­curred dur­ing the 1990 Gulf War, when the Unit­ed States used a lo­gis­tics analy­sis tool known as DART. That sys­tem re­lied on much old­er sym­bol­ic AI al­go­rithms rather than the neur­al net­works that un­der­pin to­day’s large lan­guage mod­els.

A mod­ern large lan­guage mod­el can gen­er­ate code for a lo­gis­tics analy­sis tool sim­i­lar to DART with rel­a­tive ease and as­sist mil­i­tary or­gan­i­sa­tions in im­ple­ment­ing sys­tems to stream­line their op­er­a­tions.

There is al­so the grow­ing risk that AI could be used to un­der­mine democ­ra­cy through the spread of mis­in­for­ma­tion. AI-pow­ered bots can flood so­cial me­dia with false­hoods, while AI-gen­er­at­ed videos and fab­ri­cat­ed news re­ports can be used to ma­nip­u­late pub­lic opin­ion and in­flu­ence vot­ers.

It is there­fore crit­i­cal that law­mak­ers in Trinidad and To­ba­go re­ceive reg­u­lar brief­in­gs on de­vel­op­ments in ar­ti­fi­cial in­tel­li­gence and its po­ten­tial ap­pli­ca­tions.

On the pos­i­tive side, a pow­er­ful mod­el such as Fa­ble 5 could help digi­tise en­tire gov­ern­ment min­istries with­in months. On the oth­er hand, Trinidad and To­ba­go is not ad­e­quate­ly pre­pared for the cy­ber­se­cu­ri­ty threats that in­creas­ing­ly so­phis­ti­cat­ed AI sys­tems are like­ly to pose.

The ab­sence of com­pre­hen­sive cy­ber­crime leg­is­la­tion and a well-equipped cy­ber­crime unit leaves the coun­try vul­ner­a­ble to ma­li­cious ac­tors seek­ing to weaponise ar­ti­fi­cial in­tel­li­gence.