Three months out

Cabinet Office officials warn that criminals are three to five months away from using AI to steal from bank accounts at will.

“Cyber criminals are just three months away from being able to use AI to steal at will from bank accounts”

Who is saying this

The warning comes from officials working with Britain's AI Security Institute, which tests cutting edge models before they reach the public.

They see it first

Institute insiders get early sight of the newest systems. That vantage point is why their timeline moved forward, not back.

The nightmare scenario

One official called it an AI Armageddon, and said it may arrive sooner than anyone in government previously modelled.

“this ‘AI Armageddon’ may become reality sooner than previously thought”

Guardrails you can peel off

Chinese labs favour models whose safety features can be stripped away by developers, unlike Western releases.

What guardrails actually do

They block certain words and patterns, or return a refusal, when someone asks how to build a bomb or run a scam.

Capability goes retail

As China ships more advanced models with disableable safety features, small businesses and ordinary people become reachable targets.

Every hacker gets a tool

The official's blunt version: password cracking, financial data theft, corporate data wipes and fraud, all handed to any hacker who wants them.

“Every Russian and Nigerian hacker will soon get their hands on technology which will be able to break even the most advanced passwords”

Banks are behind

The Financial Stability Institute said this week that banks can no longer respond adequately to this level of technology.

Hours, not days

A paper published last Wednesday said some attack tools are advanced enough that defenders may have only hours to shut an intruder out.

“they may only have hours to stop a hacker from breaking in”

Ministers brace for a wave

Government expects a deluge of sophisticated attacks and is worried by how fast the underlying technology is improving.

Targeted interventions

Ministers are weighing new rules for national security risks, insisting any move would be proportionate and would not choke growth.

America shuts the door

Anthropic barred Britain's safety institute from testing its latest model, straining relations between the two allies.

Treated like weapons

One AI expert read the block as proof that America now handles its models as state secrets and dangerous weapons.

“state secrets and dangerous weapons”

Five bioweapon attempts

Anthropic said it stopped five attempts this year to use its Claude model to help make bioweapons.

The same capability cuts both ways

The firm said Claude's biological research skill can drive breakthroughs, and could also help create superbugs capable of killing millions.

A researcher walks out

An Anthropic researcher quit over what he called an out of control AI arms race, saying the big labs are not acting responsibly.

More than one in ten

He put the chance of the technology killing all humans within the next decade at more than 10 per cent.

“‘more than 10 per cent’ chance of the technology killing ‘all humans’”

Speak up now

Campaigner Connor Axiotes called on officials to state their fears publicly, and criticised America for distrusting its closest ally.

“Do not be remembered as the person who failed to act.”

What you can do today

Government says firms are not defenceless: adopt Cyber Essentials, follow NCSC guidance, and put cyber security on the board agenda.

“All firms should be acting urgently to improve their cyber defences”