Phil Venables
@philvenables
Followers
14K
Following
3K
Media
2K
Statuses
9K
All about cyber, resilience, risk, AI - at scale. Partner - Ballistic Ventures / Google - Strategic Advisor / 4 x CISO / Board Director / Chief Risk Officer
USA
Joined April 2009
SUSE on Confidential Computing VMs. Defense in depth at the infrastructure layer.
cloud.google.com
Secure sensitive data on Google Cloud using SUSE Linux Enterprise Server (SLES) and Confidential VMs with AMD SEV, AMD SEV-SNP, and Intel TDX.
1
0
2
Well this is going to be amusing and not in a good way.
In a joint paper with @OwainEvans_UK as part of the Anthropic Fellows Program, we study a surprising phenomenon: subliminal learning. Language models can transmit their traits to other models, even in what appears to be meaningless data.
0
1
10
Chain of Thought is Not Explainability. Fascinating paper that, in many ways, negates the naive use of chain of thought exposition to provide hard evidence of explainability of model reasoning and output.
1
3
11
RT @joshcorman: 1yr ago… most news was RE: airline disruptions ✈️ … most of my calls were RE: patient care disruptions 🏥.
0
24
0
Great memories of this street. Mainly because after a Rugby League cup final in Edinburgh many years ago my friend Paul found a very large (like 5 foot wingspan) dead seagull there and then carried it around all the many pubs we visited that night. It was very popular.
0
0
4
Bug bounties in the cloud.
Had an amazing time at Google Cloud bugSWAT in Sunnyvale!.Got to meet and learn from some truly brilliant minds. Massive thanks to @GoogleVRP - not just the best BBP, but a team that genuinely empowers researchers, builds community, and sets the bar for the whole industry 🦄
0
3
10
RT @GoogleCloudSec: As cyber resilience emerges as a top business priority, @philvenables emphasizes why board members should expand their….
0
7
0
“Quantum mechanics is the operating system that other physical theories run on as application software.”. The whole book worth it for that one line.
0
3
9
Insurance and AI. Underwriting intelligence.
underwriting-superintelligence.com
The "Incentive Flywheel" of AI: how insurance unlocks secure Al progress and enables faster AI adoption.
1
0
5
Surprising (for many) result re the [lack of] productivity boost of AI in aspects of development. Be interesting to see comparable studies for security activities - to see which are boosted or not for what skills and circumstances.
1
1
5
This pattern repeats itself everywhere. Read more about that "uncanny valley" here.
0
0
2
Bottom line: when you are new to a role or assignment, show people this last chart and remind them that things may start to look worse before they get better and that can be a sign of being on track. Show them the point where you don’t want to be fired.
1
0
1
5. However, if you get the support from leadership and get to push through then you will hit a sustained lower level of issues. Many organizations have done this.
1
0
1
4. Then a new person turns up, applies the same method, typically reinventing/replacing what you did, and so the cycle continues. Often, given entropy and the usual net increase in risk, the graph keeps trending up despite the occasional downward move.
1
0
1
3. You start digging deeper, improve monitoring, risk assessments/instrumentation: as a result you start finding even more (previously unknown) issues that need fixing. At this point management wonders why you have made the situation worse and decide they need someone new.
1
0
1
2. You hit the ground running and find a bunch of quick wins and start reducing the number of issues. So far so good.
1
0
1