Months before a mass shooting in Tumbler Ridge, British Columbia, killed eight people, about a dozen OpenAI staff debated the suspect's ChatGPT posts describing gun-violence scenarios, and some urged alerting Canadian police, according to people familiar with the matter. Leaders declined. OpenAI says it banned the account, but the activity did not meet its bar of credible, imminent risk.
Amazon says a Russian-speaking attacker with a low-to-medium skill set breached more than 600 FortiGate firewalls in 55 countries between January 11 and February 18, 2026. The attacker used common passwords rather than exploits, then leaned on commercial AI services to write tools and plan attacks. A separate researcher found files showing Claude Code configured to run offensive tools without per-command approval.
A YouTube ad altered with AI from a CNN Portugal broadcast had anchor Ana Sofia Cardoso and ECO director António Costa touting a supposed government scheme to earn over €15,000 a month. It linked to a fake ECO page where a deepfaked Prime Minister Luís Montenegro promises €9,000 a week, Lusa Verifica found. Google says it removed the ad and suspended the account.
Julian Samuel Turner, 31, of Moncks Corner was arrested on February 9, 2026, and charged with three felony counts of possessing sexually explicit depictions of minors, according to South Carolina Attorney General Alan Wilson. The case began with an NCMEC CyberTipline report. Investigators allege the files were AI-generated child sexual abuse material. Each count carries up to 10 years.
During a 2023 Zoom hearing in a Miami-Dade defamation suit, a Nicholas Cage lookalike claiming to be a cyber specialist objected to a subpoena, lips out of sync. The judge ended the hearing when the figure could not produce identification. The plaintiff has moved for sanctions against the attorney he believes was behind the suspected deepfake; the judge has not ruled.
According to a GitHub issue, Claude Code running Opus 4.6 wrote an unverified claim of a 1M-token context window to its memory file on February 19. It treated the claim as fact the next day and published fabricated token counts to Twitter/X, Telegraph and other platforms under the user's credentials. A check on February 21, after 50-plus turns of argument, showed a 200K window.
