Story perspectives

Media Summit Pushes Digital ID, Flags AI Weaponization Risks

7/12/2026

Summary

Johannesburg Hosts Summit

The University of Johannesburg hosted the Media Summit for Government on July 8-9, 2026.
Minister Solly Malatsi called for a digitally accessible state with a trusted entry point using digital identity.
Philasande Sokhela said AI should extend intelligence, while Lorato Tshenkeng warned it could displace workers and urged ethical decision-making.
Dr Caroline Azionya, president-elect of PRISA, warned AI-generated synthetic media can be weaponised, citing a 2025 study where only 1% identified deepfakes.

Summary

Researchers Raise Accuracy

Training participants raised detection accuracy from about 40% to 80%, with a few reaching near-100% correctness.
Researchers used StyleGAN3 to create thousands of synthetic faces and taught six perceptual cues, noting AI performed worse on non-white, older or younger faces.
An Australian politician proposed that AI-generated political content be disclosed and watermarked.
A Hong Kong-based firm transferred £25m after an employee was deceived by a deep-fake video of their boss.

History

Historical Context

After Facebook and Microsoft co-hosted the 2019 Deepfake Detection Challenge, researchers produced a model that reached 65 % accuracy on a 4,000-video holdout set. The contest spurred policy moves such as the EU AI Act, the UK Online Safety Act and China’s 2023 rule requiring clear labels on synthetic media, underscoring growing governmental focus on detection standards.

1 of 3

Johannesburg Hosts Summit

The University of Johannesburg hosted the Media Summit for Government on July 8-9, 2026.
Minister Solly Malatsi called for a digitally accessible state with a trusted entry point using digital identity.
Philasande Sokhela said AI should extend intelligence, while Lorato Tshenkeng warned it could displace workers and urged ethical decision-making.
Dr Caroline Azionya, president-elect of PRISA, warned AI-generated synthetic media can be weaponised, citing a 2025 study where only 1% identified deepfakes.

1 / 3