🔍 My research on deepfake threats reveals a dangerous overconfidence gap. A lot of people I interview say: “Sure, it’s dangerous, but I’ll know.” They’re worried about protecting elderly relatives or decision-impaired individuals, but believe they can easily spot fakes themselves.
Bad actors’ use audio attacks more frequently, while video malicious deepfakes are mostly used for fun on TikTok, less for corporate fraud.
🔍 However, real advanced models can clone too well. Research shows people are better than AI—but still not very good at distinguishing deepfakes. (82% of humans beat AI in side-by-side comparison; only 13-37% beat AI in single video assessment, so AI wins.) -> https://lnkd.in/dfnjUKqP
đź’ˇ While creating deepfake video is still a hassle requiring more time, it’s much more accessible with audio—you just need 3 seconds of someone’s voice to clone it convincingly.
It’s cool that you can use deepfakes for research, movie stunts, and voice cloning for people losing their voice due to medical conditions. But bad actors will exploit its wicked functionalities.
đźš©Detection tools seem like the answer, but they’re not. NPR tested leading deepfake detectors—50% accuracy. That’s like a coin flip. -> https://lnkd.in/dq3KFedt
đźš© Deepfakes evolve faster than detection technology.
As someone who loves working remotely, I really don’t like the new operating principle offered by Jeff Crume, PhD, CISSP: “If I’m not in the room with you, I assume it’s not really you.” Brutal for remote workers.
Check the talk here: https://lnkd.in/dUDZMxNN
🤔 The solution isn’t technological—it’s behavioral.
—>Out-of-band verification before any sensitive transaction.
—>Switch communication channels.
—>Embrace healthy skepticism.
Those “clever” defenses, like asking someone to hold something in front of their face – yesterday’s solutions for tomorrow’s problems.
What’s your backup plan when you can’t trust what you see or hear?
Spoiler: Reality Defender just made their API public. I’m looking forward to testing it and hoping for better detection scores.
I got the screenshot from this video -> https://lnkd.in/dVaP_dPM
