Open Season
Open Season – A cartoon illustrating that North Korea’s army of an estimated 8,400 hackers may soon discover that loyalty to the Supreme Leader offers little protection against automation.
An indication of what could replace them comes from researchers at Wiz, where frontier models helped one researcher complete in a week what previously would have taken a world-class team months. They were finding weaknesses to fix them, but the same capabilities could help criminals exploit them, replacing individual hackers with swarms of malicious agents. This all sounds suitably scary as we start the Halloween weekend.
For their targets, that means facing attacks that can adapt and keep trying without waiting for a human hacker. Defenders need comparable automation, but accessing it through closed models can introduce obstacles. For example, testing a weakness sometimes requires attempting an exploit, which safeguards can block, and continuous investigations result in high token usage charges.
Hugging Face encountered this problem when defending against a cyberattack. Its CEO, Clément Delangue, said closed-model safeguards blocked his team, leading it to use an open model instead. Downloadable, open-weight models such as Meta’s Llama and Cisco’s Foundation-sec-8B let defenders run and adapt the technology themselves. Open models allow sensitive code to remain in-house and teams control computing capacity rather than paying an outside provider for every request.
As Delangue told the UN Security Council, “the world needs open source AI more than ever to defend itself.”
Sources:
Tim Keary (Oct 26, 2026) – Jensen Huang Says Open Models Can Beat The Frontier. Here’s How – Forbes
Thomas Brewster (Oct 23, 2026) – Inside OpenAI’s Plan To Protect The Internet Before Hackers Break It – Forbes
Shiona McCallum (Oct 24, 2026) – How one man became the unsung hero of the internet during a cyber-security scramble – BBC
