Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons
An AI model can be given a hidden backdoor by editing fewer than 10 of its rarely used neurons, without retraining the model, Microsoft researchers have shown. When a chosen trigger appears, the model outputs an attacker's command, which…
Sourcing & attribution. Newseze provides AI-curated summaries, narrative framing, and editorial analysis. The underlying reporting was contributed by The Hacker News; tap “Open original source” above to read their full reporting and support the contributing newsroom directly.
Newseze's algorithm reads the story and answers your question — calmly, factually, with source attribution. No comments, no flame wars — just answers.
No questions yet. Be the first.
Answers reflect Newseze's editorial framework applied under fair use (17 U.S.C. § 107). Not financial, legal, medical, or tax advice. Hate speech and racial slurs are blocked.
Related stories
Why it mattersAI provider keys that customers stored on Railway were stolen in an October breach and used to commit an estimated $740,000 in unauthorized spending, the platform-as-a-service provider said yesterday. It has promised to …
AI provider keys that customers stored on Railway were stolen in an October breach and used to commit an estimated $740,000 in unauthorized spending, the platfo…
By Chella·1h ago
Why it mattersMicrosoft on Tuesday released patches for a total of 1,259 vulnerabilities spanning its software portfolio, the second consecutive month where the patch volume has exceeded 1,000 flaws. These cover 726 flaws in Windows, …
Microsoft on Tuesday released patches for a total of 1,259 vulnerabilities spanning its software portfolio, the second consecutive month where the patch volume …
By Chella·1h ago
Why it mattersThe U.S. Cybersecurity and Infrastructure Security Agency (CISA) Tuesday added two security flaws impacting Cisco BroadWorks and Openwave Messaging Webtop Media Transcoder to its Known Exploited Vulnerabilities (KEV) cat…
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) Tuesday added two security flaws impacting Cisco BroadWorks and Openwave Messaging Webtop Media…
By Chella·3h ago'/> </pattern> </defs> <rect width='1200' height='675' fill='url(%23g)'/> <rect width='1200' height='675' fill='url(%23dots)'/> <rect x='60' y='210' width='6' height='80' fill='%23f5a524'/> <text x='90' y='268' font-size='15' font-weight='900' letter-spacing='5' fill='%23f5a524'>NEWSEZE</text> <g font-family='Inter, Arial, sans-serif'> <text x='60' y='320' font-size='44' font-weight='800' fill='rgba(255,255,255,0.97)' letter-spacing='-0.5'>Exploited Outlook Zero-Day Among 1,094</text><text x='60' y='376' font-size='44' font-weight='800' fill='rgba(255,255,255,0.97)' letter-spacing='-0.5'>Flaws Patched by Microsoft -</text><text x='60' y='432' font-size='44' font-weight='800' fill='rgba(255,255,255,0.97)' letter-spacing='-0.5'>SecurityWeek…</text> <text x='60' y='620' font-size='14' font-weight='700' letter-spacing='3' fill='rgba(255,255,255,0.7)'>SECURITYWEEK VIA GOOGLE NEWS · NEWSEZE</text> </g> </svg>)
Why it mattersExploited Outlook Zero-Day Among 1,094 Flaws Patched by Microsoft SecurityWeek
By Chella·14h ago