In 2023 I rated my children’s ability to reach the biscuit tin as low likelihood, high impact, and went to bed a happy man.
The tin lived on the high shelf. The small one couldn’t reach it, the middle one couldn’t climb, and the big one could probably have managed but they have a nut allergy and being the genius mastermind I am, I always say the biscuits have nuts in them.
Three children. Three lows. And that’s how a lot of risk assessments have historically been carried out.
Then one Saturday I walked into the kitchen and found a chair against the counter, the toy box balanced on the chair, and my youngest balanced on the toy box. Nobody was crying or arguing, which is how you know a crime is in progress.
The tin was empty. The crumbs were evidence. And I stood there questioning my parenting, but mostly my risk assessment.
Which reminded me of a risk assessment my friend Thom Langford and I evolved over some time and he ended up presenting it at 44CON one year as the Langford/Malik Risk Model (ver 1.0), using a pub fight to describe levels of risk and risk appetite. He put his name first. I’ve decided to be the bigger man about this and only bring it up constantly.
The way I tell it, likelihood is whether the bloke can actually land one on you. Impact is how much it hurts when he does.
It’s simple and it works. You look at the bloke, you look at his arms, whether he has any cauliflower ear and you make a call.
The trouble is that we always scored one bloke at a time.
Three weedy blokes walk in. You look at each of them individually and think, yeah, I’d fancy my chances. Low. Low. Low. Back to your mindless conversation. Nobody scored what happens when all three come at you at once, because getting three weedy blokes to co-ordinate anything takes effort, and most people are lazy.
That’s kind of what a lot of our old risk scoring was like. Each weedy little vulnerability judged on its own, found harmless, and sent home. Stringing them together was technically possible, but somebody would have to really want it.
That’s where my favourite phrase in all of security comes in. “Would require a highly sophisticated attacker.”
And I think that one phrase has kept more risks off the board than any firewall ever built. The problem is that nobody wrote it down as a control, nobody tested it, and it has now been cancelled without anybody being told.
Late in 2024, Google’s Big Sleep project, a collaboration between Project Zero and DeepMind, found an exploitable stack buffer underflow in SQLite. It got fixed the same day, before it reached an official release, and SQLite’s existing fuzzing hadn’t spotted it. Google called it the first public example of an AI agent finding a previously unknown exploitable memory-safety bug in widely used real-world software.
So the weedy bloke now has a coach. The coach sits in his ear mid-fight, has read every vulnerability write-up ever published, and has a real gift for introducing low-rated bugs to one another.
Which brings me to the zombie risk acceptance.
Somewhere in your organisation there’s a risk that was formally accepted by a man who now runs a dog grooming business in Portugal. He agreed to fight the bloke on the understanding that the bloke had had a few. That bloke has now turned up sober and brought friends.
To be fair, defenders got a coach as well. Same technology, same tireless enthusiasm. But I’ve never once seen a risk go from red to amber because somebody had a nice idea in a meeting, so I’d expect the re-scoring to go in one direction only.
Back in my kitchen, I did what any professional would do after an incident. I didn’t blame the children. I blamed the model.
