Prompt-engineering mitigations to prompt echo attacks are not robust. The only 100% reliable defense is to use a fine-tuned model, tuned without instructions in the examples. No prompt, no problems.
SECURITY
-
AI Code Generation Security Risks and Prudent Practices
By
–
I think writing `pip install …` commands from memory is bad security. If it's `python-foobar` instead of `foobar` you might be installing malware. A more prudent approach is to type `# install foobar with pip:` into Copilot and blindly execute whatever it says.
-
Have I Been Trained: Checking Your Data in AI Models
By
–
La web que comento es esta https://
haveibeentrained.com -
Cybersecurity Protection Through Collaboration Against Emerging Threats
By
–
A pleasure Knut! @JagersbergKnut thank you! And Yes it is great to see this coming together of the 'good guys' for #cybersecurity protection – absolutely what is needed when we consider the acceleration in bad actor collaboration and indeed reimagination of threats like emotet
-
DPI Consortium Launches Patent-Protected Network Security Innovation
By
–
Catalyzing #innovation in #network #Security via #patent protection New #nonprofit #DPI Consortium Launch by @NETSCOUT See http://
bit.ly/DPISecurity #CyberSecurity #DDoS #partner #education #AI @Paula_Piccard @Analytics_699 #IoT @GFCYBER @jeancayeux @JagersbergKnut #5G -
Notion’s Training Data Policy: Trust But Verify
By
–
sure, Notion says they dont use it for training in their policy https://
news.ycombinator.com/item?id=336322
29
… but none of us can know for sure unless we go audit the code ourselves -
Reverse Prompt Engineering: Notion AI Security Analysis Techniques
By
–
Prompt injection is entertaining, but harmless! https://
lspace.swyx.io/p/reverse-prom
pt-eng
… Presenting the 7 techniques I used to "reverse prompt engineer" the public alpha of Notion AI in 2 hours… and why everyone is **wrong** about the dangers of prompt injection -
Differential Privacy: Industry Funding for Theoretical AI Research
By
–
Another corollary, besides Jelani's, is that if you want to do interesting theoretical work in industry, differential privacy is a great way to trick a company into paying you to do theory (Observation attributed to an anonymous party)
-
IoT expansion to 7.6 billion devices by 2030 escalates cybersecurity risks
By
–
Edge enabled Internet of Things #IoT devices forecasted reach to 7.6 billion by 2030. These added sensors, smartphones, cameras… are exponentially growing #cybersecurity and #privacy threats and threat vectors >>> infographic @StatistaCharts via @MikeQuindazzi
-
Hacktivism Expected to Rise in 2023 Experts Warn
By
–
#Hacktivism set to rise in 2023, say experts https://
m.economictimes.com/tech/technolog
y/hacktivism-to-rise-in-2023-say-experts/amp_articleshow/96498656.cms
… #RCE #ZetoTrust #ZeroDay #cybercrime #hacker #privacy #APT #bot #CISO #DDoS #100DaysOfCode #hacking #phishing #CyberAttack #cybersecurity #Security #infosec #OpenSource #Python #javascript #rt #hackerone