Chris Vallance
speaker
69 appearances
3 recordings
1 series
first heard Jun 2026
last heard 6d ago
Chris Vallance’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 1.
Appearances
And that's easy to do in large parts of, say, the UK, but very hard in poor, remote and war-torn parts of the world and are a major reason why vaccines are wasted.
So these trials were on the tetanus diphtheria vaccine, and that's normally a liquid.
But what happened here was a sugar called tremolose was used to stabilise the vaccine as it was turned into a dry powder.
Now, some plants produce large amounts of this sugar during a drought.
They're known as resurrection plants, and they look like they have died.
But actually, the sugar has stabilised all the cells inside the plant.
And then when it rains, they come back to life.
And that's what they're
The AI Security Institute works to evaluate the capabilities of advanced AI models, but in a test of the hacking abilities of advanced models from Anthropic and OpenAI, it revealed it got rather more than it bargained for.
It said highly autonomous systems engaged in harmful activity directed at real people and organizations.
In the most alarming incident, Anthropix Mythos AI decided to use deception to carry out a hack on the online programming platform GitHub.
It created fake online profiles mimicking real people.
It fabricated evidence and sent messages in an unsuccessful attempt to pressure a real person running a programming project to include its malicious code.
When it was found out, it attempted to cover its tracks.
The AI Security Institute said another autonomous tool made by the rival company OpenAI had also attempted two less serious cyberattacks on GitHub.
Both companies said the AI Security Institute's testing had reduced or removed normal safeguards.
Anthropic said it would investigate to understand the causes of the AI's behavior.
This, in many ways, reflects a sort of real challenge for this type of AI, of generative AI, which is getting it to stick to its guardrails.
I mean, experts we've spoken to call it a game of whack-a-mole, a game of cat and mouse, if you like, in the sense that people come up with ways to jailbreak the system, to get it to break its own rules, and then the companies roll out fixes.
The AI Security Institute, that's a UK government body, found that it was able to jailbreak all the big models that it tested.
Showing 21–40 of 69 · page 2 of 4
← Previous
Next →