In brief An infosec biz has shared an audio clip of what may be a software-generated voicemail message, designed to impersonate a CEO to trick a company employee into unwittingly committing fraud.
Rumors of these so-called audio deepfake attacks have been circulating for a while. For instance, it's believed a British energy biz was scammed out of $243,000 when miscreants used algorithms to impersonate an executive’s voice on a call to convince staff to transfer cash to a crook's bank account.
It’s a short clip, handed to them by a client, of what's supposed to be a CEO leaving a voicemail for someone. In a croaky, robotic voice, the boss asks for “immediate assistance to finalize an urgent business deal." Details are vague, though we're told the recipient of the message didn't fall for it, and technical analysis of the audio showed it was likely computer generated.
Still, just letting you know.
Facebook makes a fakebook to analyze malicious behavior
Facebook has built a virtual Facebook filled with machine-learning bots to simulate and study malicious behavior on its social media platform.
The agents are designed to mimic real Facebook users; they can send messages to one another and comment on each other’s posts. The tool known as Web-Enabled Simulation (WES) was announced this month.
Now Facebook is looking for ways to use the bots for useful areas in machine learning, including multi-agent systems, game AI, and programming languages from other developers. If you’re interested in using the tech, you can submit a research proposal to get your hands on the WES platform.
Researchers said they used the software to build a fake Facebook environment to create bots looking to buy banned items on its platform, things like guns or drugs, to see whether they could bypass internal detection systems.
AI accountants that are actually, er, humans
An AI startup that boasted it had built software to automatically pay bills and file tax forms on behalf of customers has come under fire for quietly hiring humans to do most of the dirty work.
After raising $100m in a year, ScaleFactor announced it was going to shut down “a majority of its operations” starting from the end August. CEO Kurt Rathmann laid off half of his employees and blamed it all on the difficulties of operating during the COVID-19 virus pandemic.
But it appears the upstart may not have been doing too well even before the bio-nasty outbreak. Its automated accounting tools for things like bookkeeping, payroll, paying bills, and tax forms produced paperwork riddled with errors, it is claimed.
ScaleFactor even hired a human team of workers from The Outsourced Accountant in the Philippines to correct the mistakes and handle the filings, Forbes reported. One customer claimed they lost $17,000 after ScaleFactor bungled a transaction.
ScaleFactor declined to comment on the claims.
University of Florida gifted an AI supercomputer
The University of Florida will update its HiPerGator supercomputer cluster to include an Nvidia DGX SuperPOD unit, built using the chip biz's latest Ampere GPU architecture, in hopes of hitting up to 700 petaFLOPS of AI compute power.
Each DGX SuperPOD is made up of 140 DGX A100 systems, a total of 1,120 Ampere GPUs. That’s not cheap, though fortunately the university doesn’t have to spend too much money on it. Nvidia and one of its cofounders Chris Malachowsky is sending it as a $50m gift.
Malachowsky is an alumnus of the university’s Herbert Wertheim College of Engineering, and completed his degree in electrical engineering in 1980. The University of Florida will be investing another $20m to create an AI datacenter to house the supercomputer. The machine will be used to run predictive models across various subjects, including economics and medicine. The HiPerGator super is expected to be completed in 2021. ®